De Nederlandse
Mozilla-gemeenschap

“I spend 8 hours per week trying to keep up to date, it’s overwhelming!”

“Integrating new libraries is difficult. They’re either poorly maintained or updated in ways that break compatibility.”

“I want to be able to experiment quickly, without relying on APIs for closed-source models.”

These were just a few of the challenges we heard from developers during months of interviews. Today, we’re excited to introduce Blueprints and the Blueprints Hub!

Meet Mozilla.ai Blueprints

The Blueprints Hub is designed to cut through the clutter of clunky tool integration and outdated resources, so you can focus on building, not troubleshooting. It’s a showcase for the best of the open-source builder community.

What are Blueprints?

Blueprints are your go-to, customizable workflows that enable you to prototype AI applications with trusted open-source tools. Each Blueprint tackles a real-world challenge, giving you a robust foundation to work from:

• Open-source power: Fully hosted on GitHub and built with the community.
• Ready out-of-the-box: Get started instantly with accessible setup options.
• Customizable and extendable: Use it as-is or extend it to fit your own needs.
• Consistent and templatized: Every Blueprint follows a core template to keep your workflow smooth.
• Community-driven: Contribute, collaborate, and be part of something bigger.

Our launch lineup
Kick off your journey with these five practical Blueprints:

• Document-to-Podcast: Turn your text into lively, multi-voice audio clips with minimal fuss.
• Structured Question Answering: Extract answers from structured documents with a simple workflow.
• Finetuning Speech-to-text: Fine-tune speech models locally for multiple languages or your own dataset.
• OpenStreetMap AI Helper: Use computer vision to detect and map features on OpenStreetMap, with Human Verification.
• Finetuning an LLM with Federated AI: Collaboratively fine-tune models across data owners without sharing raw data.
• Build your own Timeline Algorithm: Visualize, search, and re-rank social posts using AI without data leaving your computer.

Explore the Blueprints Hub

Our new Hub is built for ease and exploration:

• Instant demos: Play around with Blueprints live in the hosted demo. No installation required.
• Video walkthroughs: Follow our video guides for a step-by-step introduction
• Technical insights: Understand the technical choices made during development of each Blueprint
• Practical use-cases: See how other developers are customizing and extending these Blueprints for their needs.
• Join our community: Share your blueprints, learn from fellow innovators, and help expand the hub.

Ready to transform your AI projects?

Join us and see how Mozilla.ai Blueprints Hub can speed up your development and spark your creativity. Visit our website now to explore, experiment, and become part of our vibrant community. Your next great idea is just a click away! 

Ready to transform your AI projects? Explore the Blueprints Hub

The post Open-source AI is hard. Blueprints can help! appeared first on The Mozilla Blog.

Sanibel Lazar is using TikTok to break beyond traditional book promotion.

Spoiler: The internet’s not finished. Welcome to “Web in Progress,” a series celebrating the internet as a space you can shape, not just scroll through. Just as Firefox empowers you to take charge of your online experience, this series highlights how individuals and communities are shaping an internet that truly serves their needs.

In this installment, see how a debut novelist is using TikTok to break beyond traditional book promotion. By focusing on niche interests, she found new ways to connect with readers who might never have picked up her book. Her experience is a testament to how digital platforms can open unexpected doors. 

Before I started promoting my debut novel, “To Have and Have More,” I hadn’t posted on any social media platform since 2015. Creating content wasn’t in my plan — until I realized it was the most practical way to get my book noticed. Working with a brand-new press meant I had to carve out my own opportunities. Social media was a means to feel like I was in the driver’s seat as my book went out into the world. 

Instead of feeling overwhelmed by the need to promote my book, I leaned into what I could control. I started creating videos on TikTok, not as part of BookTok, but tailored to themes from my book like class, privilege and wealth. They led me to unexpected audiences. I’ve ended up on PrivilegeTok, WealthTok, StatusSymbolTok and even QuietLuxuryTok — places where I stand out as the only person talking about a novel. My videos are a way to access audiences who might not otherwise pick up a book.

When one of my videos hit a couple hundred thousand views, I checked my Amazon ranking and watched my book climb. Social media has also brought me opportunities I didn’t anticipate. Rather than chasing podcast invites or op-eds, I’ve been getting invitations to do readings and guest spots (it’s thanks to social media that I got tapped to write this article) — all because people discovered me through my content. 

I’m not on TikTok to recommend books or talk about author life; instead I riff on social etiquette, classism, and luxury brands. My strategy isn’t about jumping on trend bandwagons but about getting people interested in my book. I call this approach “Oblique Content,” inspired by perfume ads that sell a mood or idea rather than focusing on product specs. In my videos, I talk about everything from toxic wealth to throwback millennial fashion trends — and I plug my book for ten seconds at the end.

I got a DM recently from a follower who said she was shocked to see a certain high-end brand at TJMaxx and thought of me. That message was a small but significant sign: My content was resonating. People were connecting my name and voice with the themes of my book.

For creatives, finding success on social media isn’t as simple as racking up views. You want your followers to be interested in your body of work and your ideas — not just your ability to “stop the scroll.” My advice? Experiment widely and don’t pigeonhole yourself in the conventions of your genre. And don’t get sidetracked scrolling for inspiration; focus on creating. 

Sanibel Lazar is a writer based in New York City. She earned her MFA from The New School and her BA in Classical Studies from the University of Pennsylvania. Her debut novel, “To Have and Have More,” will be published in April 2025. Sanibel’s work has appeared in New York Magazine, ELLE, Air Mail, Literary Hub and more.

The post Reaching readers, one TikTok at a time appeared first on The Mozilla Blog.

As you scroll through social media, the posts blend together: a heroic cat, a shocking statistic, a breaking news clip. But in a world where AI blurs the line between fact and fiction, how do you tell what’s real from what’s misinformation? 

The short answer: Most of the time, you can’t. Generative AI tools have made it faster, cheaper, and easier to create convincing fakes, flooding our feeds with AI-generated content. But here’s what you can do – learn to spot the signs of misinformation.

What should I think about when trying to detect AI? 

Just looking out for obvious AI will mean missing a lot of it. Retrain your brain to assess social media on a framework similar to the ones used by AI-misinformation researchers. Consider who’s behind the post, whether the content makes sense, how it’s written, the emotions it evokes and any signs of manipulation.   

User 

• Who is posting this? Is it a reliable source? Is this account tied to a real-world institution that I trust? 
• What is the username length? Is it a long set of random numbers and letters? Is it a verified account? Does it only have a handful of followers (who also look random or not real)? 

Content 

• Does the framing make sense? What is this content about? Is it especially vague or seems so outrageous that it couldn’t be true? Does this contradict what you already know about the topic? 
• Are there platform flags that the content could be potentially misleading, or a comment section full of claims that it’s false? Are there AI badges or indicator hashtags such as #AI, #satire, #skit, #spoof or #AD? 

Style 

• How is it written? Is there poor or wooden-sounding grammar? Is it flowery? Is there unnatural repetition, or has the user posted the same thing several times?   
• Does it repeat often-used AI words such as “elevate,” “captivate,” “tapestry” or “delve”? Does it use known AI phrases such as “provided valuable insights,” an “indelible mark,” or “a rich tapestry”? (Of course, these words and phrases don’t definitively mean that the content is AI-generated misinformation; they’re just reasons to take a closer look.)

Emotion 

• Is this an especially emotion-laded post? Is the level of emotion appropriate for the situation? 
• Does the post appear to “weaponize” emotion or tell readers how to feel about the content, such as by using more anger and swear words? (Keep in mind that bots on social media can and do use profanity). 

Manipulation 

• What might someone have to gain by touching on your emotions in this way? What’s the worst-case scenario if this turns out to not be true? What might a user (using AI) be hoping you don’t look up? 

How do I spot AI-generated images or videos online? 

Gone are the days where every AI image looked like a wacky Pixar knockoff, but it’s still worth checking for these known cues: 

• Hands and teeth with too many fingers, too many hands, too many teeth or impossibly long arms 
• Hyper-realistic images or those that look like paintings 
• Texture issues in the background, midground and around the corners and edges of an image
• Unnatural smoothness, especially on faces, clothes or hair 
• Shadows and light coming from the wrong place, or with only certain elements casting shadows 
• Abrupt transitions, either in an image or a video 

How do I verify if something is AI or not? 

Tools like TrueMedia.org scan social posts for false faces, voices, and information, while Mozilla’s Deepfake Detector analyzes text using multiple detection engines and generates a confidence score indicating the likelihood that the content is AI-generated. But while AI detection accuracy is improving, it isn’t perfect.

It always helps to try to verify the information itself — search for it along with “fact check” and look for trusted sources. For images, try a reverse image search using Google Image Search or Bing Image Match. 

What can misinformation look like on TikTok? 

Every social media site has its own AI landscape. Fake news, images and news clips targeting young voters and consumers are circulated particularly widely on TikTok due to its young user base. “Content farms” spin out inaccurate or misleading videos, often multiple a day in the distinctive TikTok style of on-screen text read by an AI voice. 

When scrolling on TikTok, be skeptical — or at least get a second opinion — on any informational videos that aren’t read by real people or only consist of captions to the AI voice (reputable news sites usually show who’s talking to build trust). Profiles that look like news sites but that have no comments or likes (especially for celebrity news) are a red flag — as well as canned phrases like “creepy response” or “finally broke their silence” meant to drive clicks. 

What can misinformation look like on X?

Though many AI-generated posts on X are largely innocuous text-based posts, videos in particular are more likely to be political deepfakes. And while the crowdsourced moderation feature, “Community Notes,” allows users to annotate posts with context or warnings, it replaced a more robust monitoring operation that means it’s more likely users will encounter bots. 

Stay wary of accounts that only spam replies, or situations where multiple accounts are commenting similar things under a post. If a user only posts replies, especially to inflammatory content, it’s a red flag that it’s a bot searching for certain keywords. 

Also, user verification on X is the least trustworthy of the major social media sites as users can pay for “verified” status (in one Cambridge study, half of synthetic profiles studied had verified status). 

What can misinformation look like on Facebook? 

It’s especially difficult to silo yourself from AI-generated content on Facebook, even if you’re only interested in posts from family and friends. Over the past three years, there has been a “significant increase” in the rate of users seeing posts they held no “friend” connection to, thanks to the algorithm that surfaces high-engagement posts in users’ feeds.

Being disciplined about clicking “not interested” under the three dots on each post can help stem the flow, as well as staying skeptical of images and being wary of link-outs to “news” sites. Verify posts (even those that appear to be from a harmless, real person) about any news events independently.

Misleading posts on Facebook are also especially focused on trying to get users off Facebook — directing them off the platform to content farms, fake stores and other scam sites. 

Stay alert and think critically online

Humans often overestimate how good they are at detecting AI — nice art is sometimes AI-generated, and terrible grammar is sometimes very human. It’s not easy to navigate a landscape designed to trick you, but your best call is to improve how you critically consume all information. Stay curious. After all, AI gets better every passing day — right down to drawing those tricky hands.

Sarah Skinner is a senior editor at a NYC tech startup. She holds a degree from Cornell University on AI and empathy, and has previously worked for McKinsey & Company and the Associated Press.

The post Misinformation in the age of AI: It’s in the details (like extra fingers) appeared first on The Mozilla Blog.

Last week the Department of Justice and some state attorneys general filed revised proposed remedies in the U.S. v. Google LLC search case. If the proposed remedies barring all search payments to browser developers are adopted by the court, these misguided plans would be a direct hit to small and independent browsers—the very forces that keep the web open, innovative and free. This case was meant to promote search competition, yet somehow the outcome threatens to crush browser competition, making it even harder for challengers to stand up to dominant players like Google, Apple and Microsoft.

“These proposed remedies prohibiting search payments to small and independent browsers miss the bigger picture—and the people who will suffer most are everyday internet users,” said Mark Surman, President of Mozilla. “Independent browsers like Firefox are on the frontlines of protecting consumer privacy, driving browser innovation, and giving people real choice on how they experience the web. But instead of promoting a fair fight, the DOJ’s remedies would tilt the playing field further into the hands of a few dominant players, diminishing consumer choice and weakening the broader internet ecosystem.”

The DOJ’s proposal hurts, not helps, browser competition

Mozilla agrees that we need to improve search competition, but the DOJ’s proposed remedies unnecessarily risk harming browser competition instead.

Here’s why:

• The DOJ wants to ban all search agreements between Google and browsers, even independent browsers that make up a smaller part of the market.
• Dominant players that own browsers, like Apple, don’t rely on search deals as they have significant revenue streams from other sources, like hardware, operating systems and app stores.
• Meanwhile, independent browsers like Firefox fund the development of their browsers mainly through search revenue––they require this revenue to survive. Search revenue underpins a large part of our work, keeping Firefox competitive and ensuring that web users have privacy-first alternatives.
• Punishing independent browsers will not solve the problem. Judge Mehta found that independent browsers account for just 1.15% of U.S. search queries. This means that cutting off our access to search deals won’t fix the issue of search dominance—not by a landslide. Instead, it hurts browser competition.

“The big unintended consequence here is the handing of power from one dominant player to another. So, from Google Search to Microsoft, or Bing for example—while shutting out the smaller, independent challengers that actually drive browser innovation and offer web users privacy and choice,” Surman added.

The last unicorn–the web can’t afford to lose Mozilla’s browser engine

Another thing missing from this conversation is something pretty important—browser engine competition.

You see, browser engines power the web. They are central to a browser’s speed, privacy and security functionality, and the browser’s ability to innovate and do things differently. But they’re very complex and require massive resources and a deep technical expertise to maintain—so much so, that right now only three major browser engines remain: Google’s Chromium, Apple’s Webkit (this engine is really only supported on Apple devices, and isn’t considered “cross-platfrom”), and then there’s Mozilla’s Gecko (which happens to be the only true cross-platform alternative to Chromium).

The DOJ’s proposal to bar search payments to independent browser developers would put Mozilla’s ability to develop and maintain Gecko at risk. If Mozilla is unable to sustain our browser engine, it would severely impact browser engine competition and mean the death of the open web as we know it—essentially, creating a web where dominant players like Google and Apple, have even more control, not less.

“This isn’t just about Firefox,” Surman explained. “If we lose our ability to maintain Gecko, it’s game over for an open, independent web. Look, Microsoft—a $3 trillion company—already gave up its browser engine in 2019 and Opera gave up theirs in 2013. If Mozilla is forced out, Google’s Chromium becomes the only cross-platform browser engine left.”

Mozilla’s role in an open web is BIGGER than our market share

Nevermind our market share, Mozilla has played an outsized role in keeping the web open, private and advocating for choice. Firefox still serves 27 million monthly active users (MAU) in the U.S. and nearly 205 million MAU globally, but our real impact comes from making the internet better by:

• Shaping the future of web standards—maintaining our own browser engine, Gecko, gives us a voice in defining how the web works and making decisions that are in support of people, not the bottom-line.
• Ensuring interoperability—we fight for a web accessible to all—where anyone can create, access, and share content seamlessly, regardless of the devices or web services they use—not locked into a few ecosystems.
• Proving that privacy-respecting technology is possible—we build critical web technologies with security, privacy and user agency at the core.

“This isn’t something we do because it’s profitable or easy,” said Surman. “We do it because it matters. The DOJ’s proposal doesn’t just miss the mark, it risks handing even more power to dominant industry players like Google or Apple, not less.”

Mozilla calls on regulators and policymakers to recognize the vital role of independent browsers and take action to nurture competition, innovation, and protect the public interest in the evolving digital landscape.

Mozilla is committed to ensuring a fair and competitive internet ecosystem, one where independent browsers can compete on a level playing field and consumers have real choice. The future of competition, innovation and the open internet depends on us.

The post Mozilla’s response to proposed remedies in U.S. v. Google appeared first on The Mozilla Blog.

In the Firefox AI Runtime, we can use multiple threads in the dedicated inference process to speed up execution times CPU. The WASM/JS environment can create a SharedArrayBuffer and run multiple threads against its content and distribute the load on several CPU cores concurrently.

Below is the time taken in seconds on a MacBook M1 Pro, which has 10 physical cores, using our PDF.js image-to-text model to generate an alt text, with different levels of concurrency:

So running several threads is a game-changer ! But using more and more threads will start to slow down execution time to a point where it will become slower than not using threads at all. 

So one question we have asked ourselves was: how can we determine the best number of threads ?

Physical vs logical cores

According to our most recent public data report, on desktop, 81% of our users are equipped with an Intel CPU, 14% with AMD and the rest are mostly Apple devices.

All modern CPUs provide more logical cores (also called “threads”) than physical cores. This happens due to technologies like Intel’s Hyper-Threading. Or AMD’s Simultaneous Multithreading (SMT).

For example, the Intel Core i9-10900K chip has 10 physical cores and 20 logical cores.

When you spin up threads equal to the number of logical cores, you might see performance gains, especially when tasks are I/O bound or if the CPU can effectively interleave instructions.

However, for compute-bound tasks (like heavy ML inference), having more threads than physical cores can lead to diminishing returns, or even performance drops, due to factors like thread scheduling overhead and cache contention.

Not all cores are created equal

On Apple Silicon, you don’t just have a quantity of cores; you have different kinds of cores. Some are high-performance cores designed for heavy lifting, while others are efficiency cores that are optimized for lower power consumption.

For instance, Apple M1 Pro chips have a combination of high-performance (8) and efficiency cores (2). The physical cores might total 10, but each performance core is designed for heavy-duty tasks, while efficiency cores typically handle background tasks that are less demanding. 

When your machine is under load with ML tasks, it’s often better to fully utilize the high-performance cores and leave some breathing room for the efficiency cores to handle background or system processes. 

Similarly, Intel’s processors have different cores, most notably starting with their 12th-generation “Alder Lake” architecture. 

These chips feature Performance-cores (P-cores) designed for demanding, single-threaded tasks, and Efficient-cores (E-cores) aimed at background or less intensive workloads. The P-cores can leverage Intel’s Hyper-Threading technology (meaning each P-core can run two logical threads), while E-cores typically run one thread each. This hybrid approach enables the CPU to optimize power usage and performance by scheduling tasks on the cores best suited for them. Like Apple Silicon’s you’d typically want to maximize utilization of the higher-performance P-cores, while leaving some headroom on the E-cores for system processes. 

Android is close to Apple Silicon’s architecture, as most devices are using ARM’s big.LITTLE (or DynamIQ) architecture – with 2 types of cores: “big” and “LITTLE”.

On mobile Qualcomm’s CPU, there can be three types: “Prime”, “Performance” and “Efficiency”. Most recently, some phones like Samsung Galaxy S24 have gained a fourth kind of core (Exynos 2400) allowing even more combinations.

To summarize, all CPU makers have cores dedicated to performance, and cores for efficiency: 

• Performance: “P-Core”, “big”, “Prime”, “Performance”
• Efficiency: “E-Core”, “LITTLE”, “Efficiency”

By combining high-efficiency and high-performance cores, Apple Silicon, Androids, and Intel based devices can strike a better balance between power consumption and raw compute power, depending on the demands of the workload.

But if you try to run all cores (performance + efficiency) at maximum capacity, you may see:

1. Less optimal thread scheduling, because tasks will bounce between slower efficiency cores and faster performance cores.
2. Contention for shared resources like the memory bus, cache.
3. And in extreme cases: thermal throttling if the system overheats, and reaches its Thermal Design Point, in which case the clock speed is throttled to cool down the system. 

This is why simply setting the thread count to “all cores, all the time” can be suboptimal for performance.

AMD on the other hand, does not have efficiency cores. Some CPUs like the Ryzen 5 8000 combine two sizes of cores Zen 4 and Zen 4c, but the latter is not an efficiency core and can also be used to run heavy-duty tasks.

navigator.hardwareConcurrency

In a browser, there is a single and simple API you can call: navigator.hardwareConcurrency

This returns the number of logical cores available. Since it’s the only API available on the web, many libraries (including the one we vendor: onnxruntime) default to using navigator.hardwareConcurrency as a baseline for concurrency.

It’s bad practice to use that value directly as it might overcommit threads as we’ve explained in the previous sections. It’s also not aware of the current system’s activity.

For that reason, ONNX formula takes the number of logical cores divided by two and will never set it higher than 4:

Math.min(4, Math.ceil((navigator.hardwareConcurrency || 1) / 2));

That formula works out ok in general, but will not take advantage of all the cores for some devices. For instance, on an Apple M1 Pro, ML tasks could use a concurrency level up to 8 cores instead of 4.

On the other end of the spectrum, a chip like Intel’s i3-1220p that we use in our CI to run tests in Windows 11, which reflects better what our users have – see our hardware section in our Firefox Public Data Report.  

It has 12 logical cores and 10 physical cores that are composed of 8 efficient cores and 2 performance cores. ONNX formula for that chip means we would run with 4 threads, where 2 would be a better value.

navigator.hardwareConcurrency is a good starting point, but it’s just a blunt instrument. It won’t always yield the true “best” concurrency for a given device and a given workload.

MLUtils.getOptimalCPUConcurrency

While it’s impossible to get the best value at any given time without considering the system activity as a whole, looking at the number of physical cores and not using “efficiency” cores, can help to get to a better value.

Llama.cpp for instance is looking at the number of physical cores to decide for concurrency, with a few twists:

• On any x86_64, it will return the number of performance cores
• On Android, and any aarch64-based devices like Apple Silicon  it will return the number of performance cores for tri-layered chips.

We’ve implemented something very similar in a C++ API that can be used via XPIDL in our inference engine:

NS_IMETHODIMP MLUtils::GetOptimalCPUConcurrency(uint8_t* _retval) { ProcessInfo processInfo = {}; if (!NS_SUCCEEDED(CollectProcessInfo(processInfo))) { return NS_ERROR_FAILURE; } #if defined(ANDROID) // On android, "big" and "medium" cpus can be used. uint8_t cpuCount = processInfo.cpuPCount + processInfo.cpuMCount; #else # ifdef __aarch64__ // On aarch64 (like macBooks) we want to avoid efficient cores and stick with "big" cpus. uint8_t cpuCount = processInfo.cpuPCount; # else // on x86_64 we're always using the number of physical cores. uint8_t cpuCount = processInfo.cpuCores; # endif #endif *_retval = cpuCount; return NS_OK; }

This function is then straightforward to use from JS shipped within Firefox to configure concurrency when we run inference:

let mlUtils = Cc["@mozilla.org/ml-utils;1"].createInstance(Ci.nsIMLUtils); const numThreads = mlUtils.getOptimalCPUConcurrency();

We’ve moved away from using navigator.hardwareConcurrency, and we’re now using this new API.

Conclusion

In our quest to find the optimal number of threads, we’re closer to reality now, but there are other factors to consider. The system will use the CPU for other applications so it’s still possible to overload it.

Using more threads is also going to use more memory in our WASM environment, which can become a real issue. Depending on the workload, each additional thread can add up to 100MiB of physical memory usage in our runtime. We’re working on reducing this overhead but on devices that don’t have a lot of memory, limiting concurrency is still our best option.

For our Firefox ML features, we are using a variety of hardware profiles in our performance CI to make sure that we try them on devices that are close to what our users have. The list of devices we have is going to grow in the next few months to make sure we cover the whole spectrum of CPUs. We’ve started collecting and aggregating metrics on a dashboard that helps us understand what can be expected when our users run our inference engine.

The hardware landscape is also evolving a lot. For example, the most recent Apple devices introduced a new instruction set, called AMX, which used to be proprietary, and gave a significant boost compared to Neon. That has now been replaced by an official API called SME. Similarly, some phones are getting more core types, which could impact how we calculate the number of cores to use. Our current algorithm could be changed the day we leverage these new APIs and hardware in our backend.

Another aspect we have not discussed in this post is using GPU or even more specialized units like NPUs, to offload our ML tasks, which will be a post on its own.

The post What is the best hardware concurrency for running inference on CPU? appeared first on The Mozilla Blog.

As we close out the Paris AI Action Summit, one thing is clear: the conversation around open and Public AI is evolving—and gaining real momentum. Just over a year ago at Bletchley Park, open source AI was framed as a risk. In Paris, we saw a major shift. There is now a growing recognition that openness isn’t just compatible with AI safety and advancing public interest AI—it’s essential to it.

We have been vocal supporters of an ecosystem grounded in open competition and trustworthy AI —one where innovation isn’t walled off by dominant players or concentrated in a single geography. Mozilla, therefore, came to this Summit with a clear and urgent message: AI must be open, human-centered, and built for the public good. And across discussions, that message resonated.

Open source AI is entering the conversation in a big way

Two particularly notable moments stood out:

• European Commission President Ursula von der Leyen spoke about Europe’s “distinctive approach to AI,” emphasizing collaborative, open-source solutions as a path forward.
• India’s Prime Minister Narendra Modi reinforced this vision, calling for open source AI systems to enhance trust and transparency, reduce bias, and democratize technology.

These aren’t just words. The investments and initiatives announced at this Summit mark a real turning point. From the launch of Current AI, an initial $400M public interest AI partnership supporting open source development, to ROOST, a new nonprofit making AI safety tools open and accessible, to the €109 billion investment in AI computing infrastructure announced by President Macron, the momentum is clear. Add to that strong signals from the EU and India, and this Summit stands out as one of the most positive and proactive international gatherings on AI so far.

At the heart of this is Public AI—the idea that we need infrastructure beyond private, purely profit-driven AI. That means building AI that serves society and promotes true innovation even when it doesn’t fit neatly into short-term business incentives. The conversations in Paris show that we’re making progress, but there’s more work to do.

Looking ahead to the next AI summit

Momentum is building, and we must forge onward. The next AI Summit in India will be a critical moment to review the progress on these announcements and ensure organizations like Mozilla—those fighting for open and Public AI infrastructure—have a seat at the table.

Mozilla is committed to turning this vision into reality—no longer a distant, abstract idea, but a movement already in motion.

A huge thanks to the organizers, partners, and global leaders driving this conversation forward. Let’s keep pushing for AI that serves humanity—not the other way around.

––Mitchell Baker
Chairwoman, Mozilla
Paris AI Action Summit Steering Committee Member

The post Paris AI Action Summit: A milestone for open and Public AI appeared first on The Mozilla Blog.

Today we want to point to one of the most exciting announcements at the Paris AI summit: the launch of ROOST, a new nonprofit to build AI safety tools for everyone. 

ROOST stands for Robust Open Online Safety Tools, and it’s solving a clear and important problem: many startups, nonprofits, and governments are trying to use AI responsibly every day but they lack access to even the most basic safety tools and resources that are available to large tech companies. This not only puts users at risk but slows down innovation. ROOST has backing from top tech companies and philanthropies alike ensuring that a broad set of stakeholders have a vested stake in its success. This is critical to building accessible, scalable and resilient safety infrastructure all of us need for the AI era. 

What does this mean practically? ROOST is building, open sourcing and maintaining modular building blocks for AI safety, and offering hands-on support by technical experts to enable organizations of all sizes to build and use AI responsibly. With that, organizations can tackle some of the biggest safety challenges such as eliminating child sexual abuse material (CSAM) from AI datasets and models. 

At Mozilla, we’re proud to have helped kickstart this work, providing a small seed grant for the research at Columbia University that eventually turned into ROOST. Why did we invest early? Because we believe the world needs nonprofit public AI organizations that at once complement and serve as a counterpoint to what’s being built inside the big commercial AI labs. ROOST is exactly this kind of organization, with the potential to create the kind of public technology infrastructure the Mozilla, Linux, and Apache foundations developed in the previous era of the internet.

Our support of ROOST is part of a bigger investment in open source AI and safety. 

In October 2023, before the AI Safety Summit in Bletchley Park, Mozilla worked with Professor Camille Francois and Columbia University to publish an open letter that stated  “when it comes to AI Safety and Security, openness is an antidote not a poison.” 

Over 1,800 leading experts and community members signed our letter, which compelled us to start the Columbia Convening series to advance the conversation around AI, openness, and safety. The second Columbia Convening (which was an official event on the road to the French AI Action Summit happening this week), brought together over 45 experts and builders in AI to advance practical approaches to AI safety. This work helped shape some of  the priorities of ROOST and create a community ready to engage with it going forward. We are thrilled to see ROOST emerge from the 100+ leading AI open source organizations we’ve been bringing together the past year. It exemplifies the principles of openness, pluralism, and practicality that unite this growing community. 

Much has changed in the last year. At the Bletchley Park summit, a number of governments and large AI labs had focused the debate on the so-called existential risks of AI — and were proposing limits on open source AI. Just 15 months later, the tide has shifted. With the world gathering at the AI Action Summit in France, countries are embracing openness as a key component of making AI safe in practical development and deployment contexts. This is an important turning point. 

ROOST launches at exactly the right time and in the right place, using this global AI summit to gather a community that will create the practical building blocks we need to enable a safer AI ecosystem. This is the type of work that makes AI safety a field that everyone can shape and improve.

The post ROOST: Open source AI safety for everyone appeared first on The Mozilla Blog.

Image generated by DALL*E

We’re shipping a new API in Firefox Nightly that will let you use our Firefox AI runtime to run offline machine learning tasks in your web extension.

Firefox AI Runtime

We’ve recently shipped a new component inside of Firefox that leverages Transformers.js (a JavaScript equivalent of Hugging Face’s Transformers Python library) and the underlying ONNX runtime engine. This component lets you run any machine learning model that is compatible with Transformers.js in the browser, with no server-side calls beyond the initial download of the models. This means Firefox can run everything on your device and avoid sending your data to third parties.

Web applications can already use Transformers.js in vanilla JavaScript, but running through our platform offers some key benefits:

• The inference runtime is executed in a dedicated, isolated process, for safety and robustness
• Model files are stored using IndexedDB and shared across origins
• Firefox-specific performance improvements are done to accelerate the runtime

This platform shipped in Firefox 133 to provide alt text for images in PDF.js, and will be used in several other places in Firefox 134 and beyond to improve the user experience.

We also want to unblock the community’s ability to experiment with these capabilities. Starting later today, developers will be able to access a new trial “ml” API in Firefox Nightly. This API is basically a thin wrapper around Firefox’s internal API, but with a few additional restrictions for user privacy and security.

There are two major differences between this API and most other WebExtensions APIs: the API is highly experimental and permission to use it must be requested after installation.

This new API is virtually guaranteed to change in the future. To help set developer expectations, the “ml” API is exposed under the “browser.trial” namespace rather than directly on the “browser” global object. Any API exposed on “browser.trial” may not be compatible across major versions of Firefox. Developers should guard against breaking changes using a combination of feature detection and strict_min_version declarations. You can see a more detailed description of how to write extensions with it in our documentation.

Running an inference task

Performing inference directly in the browser is quite exciting. We expect people will be able to build compelling features using the browser’s data locally.

Like the original Transformers that inspired it, Transformers.js uses “tasks” to abstract away implementation details for performing specific kinds of ML workloads. You can find a description of all tasks that Transformers.js supports in the project’s official documentation.

For our first iteration, Firefox exposes  the following tasks:

• text-classification – Assigning a label or class to a given text
• token-classification – Assigning a label to each token in a text
• question-answering – Retrieve the answer to a question from a given text
• fill-mask –  Masking some of the words in a sentence and predicting which words should replace those masks
• summarization – Producing a shorter version of a document while preserving its important information.
• translation – Converting text from one language to another
• text2text-generation – converting one text sequence into another text sequence
• text-generation – Producing new text by predicting the next word in a sequence
• zero-shot-classification – Classifying text into classes that are unseen during training
• image-to-text – Output text from a given image
• image-classification – Assigning a label or class to an entire image
• image-segmentation – Divides an image into segments where each pixel is mapped to an object 
• zero-shot-image-classification – Classifying images into classes that are unseen during training
• object-detection – Identify objects of certain defined classes within an image
• zero-shot-object-detection – Identify objects of classes that are unseen during training
• document-question-answering – Answering questions on document image
• image-to-image – Transforming a source image to match the characteristics of a target image or a target image domain
• depth-estimation – Predicting the depth of objects present in an image
• feature-extraction – Transforming raw data into numerical features that can be processed while preserving the information in the original dataset
• image-feature-extraction – Transforming raw data into numerical features that can be processed while preserving the information in the original image

For each task, we’ve selected a default model. See the list here EngineProcess.sys.mjs – mozsearch. These curated models are all stored in our Model Hub at https://model-hub.mozilla.org/. A Model Hub is how Hugging Face defines an online storage of models, see The Model Hub. Whether used by Firefox itself or an extension, models are automatically downloaded on the first use and cached.

Below is example below showing how to run a summarizer in your extension with the default model:

async function summarize(text) { await browser.trial.ml.createEngine({taskName: "summarization"}); const result = await browser.trial.ml.runEngine({args: [text]}); return result[0]["summary_text"]; }

If you want to use another model, you can use any model published on Hugging Face by Xenova or the Mozilla organization. For now, we’ve restricted downloading models from those two organizations, but we might relax this limitation in the future.

To use an allow-listed model from Hugging Face, you can use an options object to set the “modelHub” option to “huggingface”  and the “taskName” option to the appropriate task when creating an engine.

Let’s modify the previous example to use a model that can summarize larger texts:

async function summarize(text) { await browser.trial.ml.createEngine({ taskName: "summarization", modelHub: "huggingface", modelId: "Xenova/long-t5-tglobal-base-16384-book-summary" }); const result = await browser.trial.ml.runEngine({args: [text]}); return result[0]["summary_text"]; }

Our PDF.js alt text feature follows the same pattern:

• Gets the image to describe
• Use the “image-to-text” task with the “mozilla/distilvit” model
• Run the inference and return the generated text

This feature is built directly into Firefox, but we’ve also made a web extension example out of it, that you can find in our source code and use as a basis to build your own. See https://searchfox.org/mozilla-central/source/toolkit/components/ml/docs/extensions-api-example. For instance, it includes some code to request the relevant permission, and a model download progress bar.

We’d love to hear from you

This API is our first attempt to enable the community to build on the top of our Firefox AI Runtime. We want to make this API as simple and powerful as possible.

We believe that offering this feature to web extensions developers will help us learn and understand if and how such an API could be developed as a web standard in the future.

We’d love to hear from you and see what you are building with this.

Come and say hi in our dedicated Mozilla AI discord #firefox-ai. Discord invitation: https://discord.gg/Jmmq9mGwy7

Last but not least, we’re doing a deep dive talk at the FOSDEM in the Mozilla room Sunday February 2nd in Brussels. There will be many interesting talks in that room, see: https://fosdem.org/2025/schedule/track/mozilla/

The post Running inference in web extensions appeared first on The Mozilla Blog.

Hi, I’m Tapan. As the leader of Firefox’s Search and AI efforts, my mission is to help users find what they are looking for on the web and stay focused on what truly matters. Outside of work, I indulge my geek side by building giant Star Wars Lego sets and sharing weekly leadership insights through my blog, Building Blocks. These hobbies keep me grounded and inspired as I tackle the ever-evolving challenges of the digital world.

I’ve always been fascinated by the internet — its infinite possibilities, endless rabbit holes and the wealth of knowledge just a click away. But staying focused online can feel impossible. I spend my days solving user problems, crafting strategies, and building products that empower people to navigate the web more effectively. Yet, even I am not immune to the pull of distraction. Let me paint you a picture of my daily online life. It’s a scene many of you might recognize: dozens of tabs open, notifications popping up from every corner, and a long to-do list staring at me. In this chaos, I’ve learned that staying focused requires intention and the right tools.

Over the years, I have discovered several Firefox features that are absolute game-changers for staying productive online:

1. Pinned Tabs: Anchor your essentials

Pinned Tabs get me to my most essential tabs in one click. I have a few persistent pinned tabs — my email, calendar, and files — and a few “daily” pinned tabs — my “must-dos” of the day tabs. This is my secret weapon for keeping my workspace organized. Pinned Tabs stay put and don’t clutter my tab bar, making it easy to switch between key resources without hunting my tab list.

To pin a tab, right-click it and select “Pin Tab.” Now, your essential tabs will always be at your fingertips.

2. Search: Use the fast lane

The “@” shortcut is my productivity superpower, taking me to search results in a flash. By typing “@amazon,” “@bing,” or “@history” followed by your search terms, you can instantly search those platforms or your browsing history without leaving your current page. This saves me time by letting me jump right to search results.

In the next Firefox update, we are making the search term persistent in the address bar so that you can use the address bar to refine your searches for supported sites.

To search supported sites, type “@” in the address bar and pick any engine from the supported list.

3. AI-powered summarization: Cut to the chase

This is one of my favorite recent additions to Firefox. Our AI summarization feature can distill long articles or documents into concise summaries, helping you grasp the key points without wading through endless text. Recently, I used Firefox’s AI summarization to condense sections of research papers on AI. This helped me quickly grasp the key findings and apply them to our strategy discussions for enhancing Firefox’s AI features. Using AI to help build AI!

To use AI-powered summarization, type “about:preferences#experimental” in the address bar and enable “AI chatbot.” Pick your favorite chatbot and sign in. Select any text on a page you wish to summarize and right-click to pick “Ask <your chatbot>.” We are adding new capabilities to this list with every release.

4. Close Duplicate Tabs: Declutter your workspace

If you are like me, you’ve probably opened the same webpage multiple times without realizing it. Firefox’s “Close Duplicate Tabs” feature eliminates this problem.

By clicking the tab list icon  at the top-right corner of the Firefox window, you can detect and close duplicate tabs, keeping your workspace clean and reducing mental load. This small but mighty tool is for anyone prone to tab overload.

5. Reader View: Eliminate distractions

Reader View transforms cluttered web pages into clean, distraction-free layouts. You can focus entirely on the content by stripping away ads, pop-ups, and other distractions. Whether reading an article or researching, this feature keeps your mind on the task.

To enable it, click the Reader View icon in the address bar when viewing a page.

These Firefox features have transformed how I navigate the web, helping me stay focused, productive, and in control of my time online. Whether managing a complex task, diving into research, or just trying to stay on top of your daily tasks, these tools can help you take charge of your browsing experience.

What are your favorite Firefox productivity tips? I would love to hear how you customize Firefox to fit your life.

Let’s make the web work for us!

Get Firefox Get the browser that protects what’s important

The post Supercharge your day: Firefox features for peak productivity appeared first on The Mozilla Blog.

Participants of the Dataset Convening in Amsterdam. Update: Following the 2024 Mozilla AI Dataset Convening, AI builders and researchers publish best practices for creating open datasets for LLM training. 

Training datasets behind large language models (LLMs) often lack transparency, a research paper published by Mozilla and EleutherAI explores how openly licensed datasets that are responsibly curated and governed can make the AI ecosystem more equitable. The study is co-authored with thirty leading scholars and practitioners from prominent open source AI startups, nonprofit AI labs, and civil society organizations who attended the Dataset Convening on open AI datasets in June 2024.

Many AI companies rely on data crawled from the web, frequently without the explicit permission of copyright holders. While some jurisdictions like the EU and Japan permit this under specific conditions, the legal landscape in the United States remains murky. This lack of clarity has led to lawsuits and a trend toward secrecy in dataset practices—stifling transparency, accountability, and limiting innovation to those who can afford it.

For AI to truly benefit society, it must be built on foundations of transparency, fairness, and accountability—starting with the most foundational building block that powers it: data. 

The research, “Towards Best Practices for Open Datasets for LLM Training,” outlines possible tiers of openness, normative principles, and technical best practices for sourcing, processing, governing, and releasing open datasets for LLM training, as well as opportunities for policy and technical investments to help the emerging community overcome its challenges. 

READ THE RESEARCH HERE

Building toward a responsible AI future requires collaboration across legal, technical, and policy domains, along with investments in metadata standards, digitization, and fostering a culture of openness. 

To help advance the field, the paper compiles best practices for LLM builders, including guidance on Encoding preferences in metadata, Data sourcing, Data Processing, Data Governance/Release, and Terms of Use.

To explore the recommendations check the full paper (also available on arXiv). 

We are grateful to our collaborators – 273 Ventures, Ada Lovelace Institute, Alan Turing Institute, Cohere For AI, Common Voice, Creative Commons, Data Nutrition Project, Data Provenance Initiative, First Languages AI Reality (Mila), Gretel, HuggingFace, LLM360, Library Innovation Lab (Harvard), Open Future, Pleias, Spawning, The Distributed AI Research Institute, Together AI, and Ushahidi– for their leadership in this work, as well as Computer Says Maybe for their facilitation support. 

We look forward to the conversations it will spark.

Previous post published on July 2, 2024:

Mozilla and EleutherAI brought together experts to discuss a critical question: How do we create openly licensed and open-access LLM training datasets and how do we tackle the challenges faced by their builders?

On June 11, on the eve of MozFest House in Amsterdam, Mozilla and EleutherAI convened an exclusive group of 30 leading scholars and practitioners from prominent open-source AI startups, nonprofit AI labs and civil society organizations to discuss emerging practices for a new focus within the open LLM community: creating open-access and openly licensed LLM training datasets. 

This work is timely. Although sharing training datasets was once common practice among many AI actors, increased competitive pressures and legal risks have made it almost unheard of nowadays for pre-training datasets to be shared or even described by their developers. However, just as open-source software has made the internet safer and more robust, we at Mozilla and EleutherAI believe open-access data is a public good that can empower developers worldwide to build upon each other’s work. It fosters competition, innovation and transparency, providing clarity around legal standing and an ability to stand up to scrutiny.

Leading AI companies want us to believe that training performant LLMs without copyrighted material is impossible. We refuse to believe this. An emerging ecosystem of open LLM developers have created LLM training datasets —such as Common Corpus, YouTube-Commons, Fine Web, Dolma, Aya, Red Pajama and many more—that could provide blueprints for more transparent and responsible AI progress. We were excited to invite many of them to join us in Amsterdam for a series of discussions about the challenges and opportunities of building an alternative to the current status quo that is open, legally compliant and just. 
During the event, we drew on the learnings from assembling “Common Pile” (the soon-to-be-released dataset by EleutherAI composed only of openly licensed and public domain data) which incorporates many learnings from its hugely successful predecessor, “The Pile.” At the event, EleutherAI released a technical briefing and an invitation to public consultation on Common Pile.

Participants engaged in a discussion at “The Dataset Convening,” hosted by Mozilla and EleutherAI on June 11, 2024 to explore creating open-access and openly licensed LLM training datasets.

Our goal with the convening was to bring in the experiences of open dataset builders to develop normative and technical recommendations and best practices around openly licensed and open-access datasets. Below are some highlights of our discussion:

• Openness alone does not guarantee legal compliance or ethical outcomes, we asked which decision points can contribute to datasets being more just and sustainable in terms of public good and data rights. 
• We discussed what “good” looks like, what we want to avoid, what is realistic and what is already being implemented in the realm of sourcing, curating, governing and releasing open training datasets. 
• Issues such as the cumbersome nature of sourcing public domain and openly licensed data (e.g. extracting text from PDFs), manual verification of metadata, legal status of data across jurisdictions, retractability of consent, preference signaling, reproducibility and data curation and filtering were recurring themes in almost every discussion.
• To enable more builders to develop open datasets and unblock the ecosystem, we need financial sustainability and smart infrastructural investments that can unblock the ecosystem.
• The challenges faced by open datasets today bear a resemblance to those encountered in the early days of open source software (data quality, standardization and sustainability). Back then, it was the common artifacts that united the community and provided some shared understanding and language. We saw the Dataset Convening as an opportunity to start exactly there and create shared reference points that, even if not perfect, will guide us in a common direction.
• The final insight round underscored that we have much to learn from each other: we are still in the early days of solving this immense challenge, and this nascent community needs to collaborate and think in radical and bold ways.

Participants at the Mozilla and EleutherAI event collaborating on best practices for creating open-access and openly licensed LLM training datasets.

We are immensely grateful to the participants in the Dataset Convening (including some remote contributors):

• Stefan Baack — Researcher and Data Analyst, Insights, Mozilla
• Mitchell Baker — Chairwoman, Mozilla Foundation
• Ayah Bdeir — Senior Advisor, Mozilla
• Julie Belião — Senior Director of Product Innovation, Mozilla.ai
• Jillian Bommarito — Chief Risk Officer, 273 Ventures
• Kasia Chmielinski — Project Lead, Data Nutrition Project
• Jennifer Ding — Senior Researcher, Alan Turing Institute
• Alix Dunn — CEO, Computer Says Maybe
• Marzieh Fadaee — Senior Research Scientist, Cohere For AI
• Maximilian Gahntz — AI Policy Lead, Mozilla
• Paul Keller — Director of Policy and Co-Founder, Open Future
• Hynek Kydlíček — Machine Learning Engineer, HuggingFace
• Pierre-Carl Langlais — Co-Founder, Pleias
• Greg Leppert — Director of Product and Research, the Library Innovation Lab, Harvard
• EM Lewis-Jong — Director, Common Voice, Mozilla
• Shayne Longpre — Project Lead, Data Provenance Initiative
• Angela Lungati — Executive Director, Ushahidi
• Sebastian Majstorovic — Open Data Specialist, EleutherAI
• Cullen Miller — Vice President of Policy, Spawning
• Victor Miller — Senior Product Manager, LLM360
• Kasia Odrozek — Director, Insights, Mozilla
• Guilherme Penedo — Machine Learning Research Engineer, HuggingFace
• Neha Ravella — Research Project Manager, Insights Mozilla
• Michael Running Wolf — Co-Founder and Lead Architect, First Languages AI Reality, Mila
• Max Ryabinin — Distinguished Research Scientist, Together AI 
• Kat Siminyu — Researcher, The Distributed AI Research Institute
• Aviya Skowron — Head of Policy and Ethics, EleutherAI
• Andrew Strait — Associate Director, Ada Lovelace Institute
• Mark Surman — President, Mozilla Foundation
• Anna Tumadóttir — CEO, Creative Commons
• Marteen Van Segbroeck — Head of Applied Science, Gretel
• Leandro von Werra — Chief Loss Officer, HuggingFace
• Maurice Weber — AI Researcher, Together AI
• Lee White — Senior Full Stack Developer, Ushahidi
• Thomas Wolf — Chief Science Officer and Co-Founder, HuggingFace

In the coming weeks, we will be working with the participants to develop common artifacts that will be released to the community, along with an accompanying paper. These resources will help researchers and practitioners navigate the definitional and executional complexities of advancing open-access and openly licensed datasets and strengthen the sense of community. 

The event was part of the Mozilla Convening Series, where we bring together leading innovators in open source AI to tackle thorny issues and help move the community and movement forward. Our first convening was the Columbia Convening where we invited 40 leading scholars and practitioners to develop a framework for defining what openness means in AI. We are committed to continuing the efforts to support communities invested in openness around AI and look forward to helping grow and strengthen this movement. 

The post Mozilla, EleutherAI publish research on open datasets for LLM training appeared first on The Mozilla Blog.

As a student pursuing a master’s degree, I’ve spent too much time searching for PDF editors to fill out forms, take notes and complete projects. I discovered Firefox’s built-in PDF editor while interning at Mozilla as a corporate communications intern. No more giving out my email address or downloading dubious software, which often risks data. The built-in PDF tool on Firefox is a secure, efficient solution that saves me time. Here’s how it has made my academic life easier. 

Fill out applications and forms effortlessly

Remember those days when you had to print a form, fill it out and then scan it back into your computer? I know, tedious. With Firefox’s PDF editor, you can fill in forms online directly from your browser. Just open the PDF in Firefox on your smartphone or computer, click the “text” button, and you’re all set to type away. It’s a gamechanger for all those scholarship applications and administrative forms, or even adult-life documents we consistently have to fill. 

Highlight and annotate lecture slides for efficient note-taking

I used to print my professors’ lecture slides and study materials just to add notes. Now, I keep my annotations within the browser – highlighting key points and adding notes. You can even choose your text size and color. This capability not only enhances my note-taking, it saves some trees too. No more losing 50-page printed slides around campus. 

Sign documents electronically without hassle

Signing a PDF document was the single biggest dread I had as a millennial, a simple task made difficult. I used to have to search “free PDF editor” online, giving my personal information to make an account in order to use free software. Firefox makes it simple. Here’s how: Click the draw icon, select your preferred color and thickness, and draw directly on the document. Signing documents electronically finally feels like a 21st century achievement. 

Easily insert and customize images in your PDFs

Sometimes, adding an image to your PDF is necessary, whether it’s a graph for a report or a picture for a project. Firefox lets you upload and adjust images right within the PDF. You can even add alternative text or alt-text to make your documents more accessible, ensuring everyone in your group can understand your work.

There are endless ways to make Firefox your own, however you choose to navigate the internet. We want to know how you customize Firefox. Let us know and tag us on X or Instagram at @Firefox.

Get Firefox Get the browser that protects what’s important

The post Streamline your schoolwork with Firefox’s PDF editor appeared first on The Mozilla Blog.

Continuing our series on Anonym’s technology, this post focuses on Anonym’s use of differential privacy. Differential privacy is a cornerstone of Anonym’s approach to building confidential and effective data solutions. In this post, we’ll explain why we integrate differential privacy (DP) into all our systems and share how we tailor our implementation to meet the unique demands of advertising use cases.

As a reminder, Mozilla acquired Anonym over the summer of 2024, as a key pillar in its effort to raise the standards of privacy in the advertising industry. Separate from Mozilla surfaces like Firefox, which work to protect users from excessive data collection, Anonym provides ad tech infrastructure that focuses on improving privacy and limiting data shared between advertisers and ad platforms. 

What is differential privacy?

 Created in 2006 by Cynthia Dwork and her collaborators, DP provides a principled method to generate insights without compromising individual confidentiality. This is typically achieved by adding carefully calibrated statistical noise to computations, making individual data points indistinguishable. 

Differential Privacy has been used in a number of different contexts to enhance user privacy, notably in the US Census  and for public health use cases. This post will focus on why Anonym believes DP is an essential tool in how we create performance with our partners, while preserving privacy. For those interested in learning more about the theoretical underpinnings of DP, we’ve linked some of our favorite resources at the end of this post.

Why differential privacy for advertising use cases?

Simply put, we believe that differential privacy offers improved privacy to users while allowing analysis on ad performance. Many traditional privacy techniques used in advertising are at high risk of exposing user data, even if inadvertently. One of the most common traditional techniques is only returning aggregates when more than a minimum number of users have contributed (thresholding). The two examples below illustrate where thresholding can still result in revealing user data. 

Example 1:  In attribution reporting, measuring partially overlapping groups can reveal individual user information. Imagine a dataset that provides attribution data segmented by age group and we have implemented a threshold of ten – meaning we will only provide reporting if we have at least ten conversions for the segment. Suppose there are only nine purchasers in the “18-20” age group. Thresholding might suppress this entire segment to protect privacy. However, if a larger group—such as users exposed to ads targeted at users aged 18 to 35—is reported, and this larger group contains just one more user, it becomes relatively straightforward to deduce that the additional user is a purchaser. This demonstrates how thresholding alone can unintentionally expose individual data by leaving related groups visible. 

Example 2: Imagine a clean room consistently suppresses results for aggregations with fewer than ten individuals but always reports statistics for groups with ten or more, an attacker could introduce minor changes to the input data—such as adding a single individual—to observe how the output changes. By monitoring these changes, the attacker could reverse-engineer the behavior of the individual added.

The FTC has recently shared its perspective that relying purely on confidential computing by using data clean rooms may not adequately protect people’s privacy and we agree – users need more protection than afforded by simple aggregation and thresholding.

The advantages of differential privacy

Differential privacy offers several key improvements over the methods discussed above:

1. Mathematical guarantees: Differential privacy provides quantifiable and provable mathematical guarantees about the confidentiality of individuals in a dataset, ensuring that the risk of revealing individual-level information is reduced. Additionally DP has a concept called composibility which states that even if we look at a large number of results over time, we can still quantify the privacy. 
2. Protection from auxiliary information: DP ensures that even if a party such as an ad platform possesses additional information about users (which is typically the case), they cannot confidently identify specific individuals from the dataset.
3. Minimal impact on utility: When implemented well, the actionability of DP-protected outputs is comparable to results without DP, and there is no need to suppress results. This means advertisers can trust their data to inform decision-making without compromising individual user confidentiality.

With these benefits, DP offers better privacy guarantees than other methods. We don’t need to think through all the potential edge cases like we saw for thresholding. For advertisers and platforms, the choice is clear: why wouldn’t you want the strongest available privacy protection?

How Anonym implements differential privacy

At Anonym, we recognize that one-size-fits-all solutions rarely work, especially in the complex world of advertising. That’s why all our DP implementations are bespoke to the ad platform and designed to maximize utility for each of their advertiser use cases.

Tailoring DP to the problem

Our approach takes into account the unique requirements of each advertising campaign. We use differential privacy for our ML-based solutions, but let’s use a measurement example:

• Measurement goals: Are we measuring the number of purchases, the amount purchased, or both? We only want to release the necessary information to maximize utility.  
• Decision context: What metrics matter most to the advertiser? In lift that could be understanding incrementality vs. statistical significance. We can tailor what we return to meet the advertiser’s needs. This increases utility by avoiding releasing information that will not change decision making.  
• Dimensional Complexity: What dimensions are we trying to measure? Is there a hierarchy? We can improve utility by taking advantage of underlying data structures.

High utility DP requires expertise

To create solutions that are both private and actionable, our development process involves close collaboration between our teams of differential privacy experts and advertising experts.

Differential privacy experts play a crucial role in ensuring the mathematical correctness of implementations. This is a critical step because DP guarantees are only valid if implemented correctly. These DP experts carefully match the DP method to the specific problem, selecting the option that offers the highest utility. Additionally, these experts incorporate the latest innovations in DP to further enhance the effectiveness and practicality of the solutions.

Advertising experts, on the other hand, help ensure the base ads algorithms are optimized to deliver high-utility results. Their insights further optimize DP methods for decision-making, aligning the outputs with the specific needs of advertisers.

This multidisciplinary approach helps our solutions meet rigorous mathematical privacy standards while empowering advertisers to make effective, data-driven decisions.

Conclusion

In an era of increasing data collection and heightened privacy concerns, differential privacy is a key technique for protecting the confidentiality of individual data without sacrificing utility. At Anonym, we’ve built DP into the foundation of our systems because we believe it’s the best way to deliver actionable insights while safeguarding user trust.

By combining deep expertise in DP with a nuanced understanding of advertising, we’re able to offer solutions that meet the needs of advertisers, regulators, and, most importantly, people.

Further Reading: Check out our favorite resources to learn more about differential privacy:

• Programming Differential Privacy 
• This blog series
• Open DP’s resources (here) 

The post Raising the bar: Why differential privacy is at the core of Anonym’s approach appeared first on The Mozilla Blog.

On Nov. 19, 2024, Mozilla and Columbia University’s Institute of Global Politics held the Columbia Convening on AI Openness and Safety in San Francisco. The Convening, which is an official event on the road to the AI Action Summit to be held in France in February 2025, took place on the eve of the Convening of the International Network of AI Safety Institutes. In the convening we brought together over 45 experts and practitioners in AI to advance practical approaches to AI safety that embody the values of openness, transparency, community-centeredness and pragmatism. 

Prior to the event on Nov. 19, twelve of these experts formed our working group and collaborated over six weeks on a thorough, 40-page “backgrounder” document that helped frame and focus our-person discussions, and design tracks for participants to engage with throughout the convening. 

The Convening explored the intersection of Open Source AI and Safety, recognizing two key dynamics. First, while the open source AI ecosystem continues to gain unprecedented momentum among practitioners, it seeks more open and interoperable tools to ensure responsible and trustworthy AI deployments. Second, this community is approaching safety systems and tools differently, favoring open source values that are decentralized, pluralistic, culturally and linguistically diverse, and emphasizing transparency and auditability. Our discussions resulted in a concrete, collective and collaborative output: “A Research Agenda for a Different AI Safety,” which is organized around five working tracks.

We’re grateful to the French Government’s AI Action Summit for co-sponsoring our event as a critical milestone on the “Road to the AI Action Summit” in February, and to the French Minister for Artificial Intelligence who joined us to give closing remarks at the end of the day. 

In the coming months, we will publish the proceedings of the conference. In the meantime, a summarized readout of the discussions from the convening are provided below. 

Readout from Convening: What’s missing from taxonomies of harm and safety definitions?

Participants grappled with the premise that there is no such thing as a universally ‘aligned’ or ‘safe’ model. We explored the ways that collective input can both support better-functioning AI systems across use cases, help prevent harmful uses of AI systems, and further develop levers of accountability.  Most AI safety challenges involve complex sociotechnical systems where critical information is distributed across stakeholders and key actors often have conflicts of interest, but participants noted that open and participatory approaches can help build trust and advance human agency amidst these interconnected and often exclusionary systems. 

Participants examined limitations in existing taxonomies of harms and explored what notions of safety put forth by governments and big tech companies can fail to capture. AI-related harms are often narrowly defined by companies and developers for practical reasons, who often overlook or de-emphasize broader systemic and societal impacts on the path to product launches. The Convening’s discussions emphasized that safety cannot be adequately addressed without considering domain-specific contexts, use cases, assumptions, and stakeholders. From automated inequality in public benefits systems to algorithmic warfare, discussions highlighted how safety discussions accompanying AI systems’ deployments can become too abstract and fail to center diverse voices and the individuals  and communities who are actually harmed by AI systems. A key takeaway was to continue to ensure AI safety frameworks center human and environmental welfare, rather than predominantly corporate risk reduction. Participants also emphasized that we cannot credibly talk about AI safety without acknowledging the use of AI in warfare and critical systems, especially as there are present day harms playing out in various parts of the world.

Drawing inspiration from other safety-critical fields like bioengineering, healthcare, and public health, and lessons learned from adjacent discipline of Trust and Safety, the workshop proposed targeted approaches to expand AI safety research. Recommendations included developing use-case-specific frameworks to identify relevant hazards, defining stricter accountability standards, and creating clearer mechanisms for harm redressal. 

Safety tooling in open AI stacks

As the ecosystem of open source tools for AI safety continues to grow, developers need better ways to navigate it. Participants mapped current technical interventions and related tooling, and helped identify gaps to be filled for safer systems deployments. We discussed the need for reliable safety tools, especially as post-training models and reinforcement learning continues to evolve. Conversants noted that high deployment costs, lack of safety tooling and methods expertise, and fragmented benchmarks can also hinder safety progress in the open AI space. Resources envisioned included dynamic, standardized evaluations, ensemble evaluations, and readily available open data sets that could help ensure that safety tools and infrastructure remain relevant, useful, and accessible for developers. A shared aspiration emerged: to expand access to AI evaluations while also building trust through transparency and open-source practices.

Regulatory and incentive structures also featured prominently, as participants emphasized the need for clearer guidelines, policies, and cross-sector alignment on safety standards. The conversation noted that startups and larger corporations often approach AI safety differently due to contrasting risk exposures and resourcing realities, yet both groups need effective monitoring tools and ecosystem support. The participants explored how insufficient taxonomical standards, lack of tooling for data collection, and haphazard assessment frameworks for AI systems can hinder progress and proposed collaborative efforts between governments, companies, and non-profits to foster a robust AI safety culture. Collectively, participants envisioned a future where AI safety systems compete on quality as much as AI models themselves.

The future of content safety classifiers

AI systems developers often have a hard time finding the right content safety classifier for their specific use case and modality, especially when developers need to also fulfill other requirements around desired model behaviors, latency, performance needs, and other considerations. Developers need a better approach for standardizing reporting about classifier efficacy, and for facilitating comparisons to best suit their needs. The current lack of an open and standardized evaluation mechanism across various types of content or languages can also lead to unknown performance issues, requiring developers to perform a series of time-consuming evaluations themselves — adding additional friction to incorporating safety practices into their AI use cases.

Participants chartered a future roadmap for open safety systems based on open source content safety classifiers, defining key questions, estimating necessary resources, and articulating research agenda requirements while drawing insights from past and current classifier system deployments. We explored gaps in the content safety filtering ecosystem, considering both developer needs and future technological developments. Participants paid special attention to the challenges posed in combating child sexual abuse material and identifying other harmful content. We also noted the limiting factors and frequently Western-centric nature of current tools and datasets for this purpose, emphasizing the need for multilingual, flexible, and open-source solutions. Discussions also called for resources that are accessible to developers across diverse skill levels, such as a “cookbook” offering practical steps for implementing and evaluating classifiers based on specific safety priorities, including child safety and compliance with international regulations.

The workshop underscored the importance of inclusive data practices, urging a shift from rigid frameworks to adaptable systems that cater to various cultural and contextual needs and realities. Proposals included a central hub for open-source resources, best practices, and evaluation metrics, alongside tools for policymakers to develop feasible guidelines. Participants showed how AI innovation and safety could be advanced together, prioritizing a global approach to AI development that works in underrepresented languages and regions.

Agentic risk

With growing interest in “agentic applications,” participants discussed how to craft meaningful working definitions and mappings of the specific needs of AI-system developers in developing safe agentic systems. When considering agentic AI systems, many of the usual risk mitigation approaches for generative AI systems — such as content filtering or model tuning —  run into limitations. In particular, such approaches are often focused on non-agentic systems that only generate text or images, whereas agentic AI systems take real-world actions that carry potentially significant downstream consequences. For example, an agent might autonomously book travel, file pull requests on complex code bases, or even take arbitrary actions on the web, introducing new layers of safety complexity. Agent safety can present a fundamentally different challenge as agents perform actions that may appear benign on their own while potentially leading to unintended or harmful consequences when combined.

Discussions began with a foundational question: how much trust should humans place in agents capable of decision-making and action? Through case studies that included AI agents being used to select a babysitter and book a vacation, participants analyzed risks including privacy leaks, financial mismanagement, and misalignment of objectives. A clear distinction emerged between safety and reliability; while reliability errors in traditional AI might be inconveniences, errors in autonomous agents could cause more direct, tangible, and irreversible harm. Conversations highlighted the complexity of mitigating risks such as data misuse, systemic bias, and unanticipated agent interactions, underscoring the need for robust safeguards and frameworks.

Participants proposed actionable solutions focusing on building transparent systems, defining liability, and ensuring human oversight. Guardrails for both general-purpose and specialized agents, including context-sensitive human intervention thresholds and enhanced user preference elicitation, were also discussed. The group emphasized the importance of centralized safety standards and a taxonomy of agent actions to prevent misuse and ensure ethical behavior. With the increasing presence of AI agents in sectors like customer service, cybersecurity, and administration, Convening members stressed the urgency of this work.

Participatory inputs

Participants examined how participatory inputs and democratic engagement can support safety tools and systems throughout development and deployment pipelines, making them more pluralistic and better adapted to specific communities and contexts. Key concepts included creating sustainable structures for data contribution, incentivizing safety in AI development, and integrating underrepresented voices, such as communities in the Global Majority. Participants highlighted the importance of dynamic models and annotation systems that balance intrinsic motivation with tangible rewards. The discussions also emphasized the need for common standards in data provenance, informed consent, and participatory research, while addressing global and local harms throughout AI systems’ lifecycles.

Actionable interventions such as fostering community-driven AI initiatives, improving tools for consent management, and creating adaptive evaluations to measure AI robustness were identified. The conversation called for focusing on democratizing data governance by involving public stakeholders and neglected communities, ensuring data transparency, and avoiding “golden paths” that favor select entities. The workshop also underscored the importance of regulatory frameworks, standardized metrics, and collaborative efforts for AI safety.

Additional discussion

Some participants discussed the tradeoffs and false narratives embedded in the conversations around open source AI and national security. A particular emphasis was placed on the present harms and risks from AI’s use in military applications, where participants stressed that these AI applications cannot solely be viewed as policy or national security issues, but must also be viewed as technical issues too given key challenges and uncertainties around safety thresholds and system performance.

Conclusion

Overall, the Convening advanced discussions in a manner that showed that a pluralistic, collaborative approach to AI safety is not only possible, but also necessary. It showed that leading AI experts and practitioners can bring much needed perspectives to a debate dominated by large corporate and government actors, and demonstrated the importance of a broader range of expertise and incentives. This framing will help ground a more extensive report on AI safety that will follow from this Convening in the coming months.

We are immensely grateful to the participants in the Columbia Convening on AI Safety and Openness; as well as our incredible facilitator Alix Dunn from Computer Says Maybe, who continues to support our community in finding alignment around important socio-technical topics at the intersection of AI and Openness.

The list of participants at the Columbia Convening is below, individuals with an asterisk were members of the working group 

• Guillaume Avrin – National Coordinator for Artificial Intelligence, Direction Générale des Entreprises
• Adrien Basdevant – Tech Lawyer, Entropy
• Ayah Bdeir* – Senior Advisor, Mozilla
• Brian Behlendorf – Chief AI Strategist, The Linux Foundation 
• Stella Biderman– Executive Director, EleutherAI 
• Abeba Birhane – Adjunct assistant professor, Trinity College Dublin 
• Rishi Bommasani – Society Lead, Stanford CRFM
• Herbie Bradley – PhD Student, University of Cambridge
• Joel Burke – Senior Policy Analyst, Mozilla 
• Eli Chen – CTO & Co-Founder, Credo AI
• Julia DeCook, PhD – Senior Policy Specialist, Mozilla 
• Leon Derczynski – Principal research scientist, NVIDIA Corp & Associate professor, IT University of Copenhagen
• Chris DiBona – Advisor, Unaffiliated
• Jennifer Ding – Senior researcher, The Alan Turing Institute 
• Bonaventure F. P. Dossou – PhD Student, McGill University/Mila Quebec AI Institute 
• Alix Dunn – Facilitator, Computer Says Maybe 
• Nouha Dziri* – Head of AI Safety, Allen Institute for AI 
• Camille François* – Associate Professor, Columbia University’s School of International and Public Affairs
• Krishna Gade – Founder & CEO, Fiddler AI 
• Will Hawkins* – PM Lead for Responsible AI, Google DeepMind 
• Ariel Herbert-Voss – Founder and CEO, RunSybil 
• Sara Hooker – VP Research, Head of C4AI, Cohere
• Yacine Jernite* – Head of ML and Society, HuggingFace 
• Sayash Kapoor* – Ph.D. candidate, Princeton Center for Information Technology Policy
• Heidy Khlaaf* – Chief AI Scientist, AI Now Institute 
• Kevin Klyman – AI Policy Researcher, Stanford HAI 
• David Krueger – Assistant Professor, University of Montreal / Mila 
• Greg Lindahl – CTO, Common Crawl Foundation
• Yifan Mai – Research Engineer, Stanford Center for Research on Foundation Models (CRFM)
• Nik Marda* – Technical Lead, AI Governance, Mozilla
• Petter Mattson – President, ML Commons 
• Huu Nguyen – Co-founder, Partnership Advocate, Ontocord.ai 
• Mahesh Pasupuleti – Engineering Manager, Gen AI, Meta 
• Marie Pellat* – Lead Applied Science & Safety, Mistral 
• Ludovic Péran* – AI Product Manager
• Deb Raji* – Mozilla Fellow 
• Robert Reich – Senior Advisor, U.S. Artificial Intelligence Safety Institute
• Sarah Schwetmann – Co-Founder, Transluce & Research Scientist, MIT
• Mohamed El Amine Seddik – Lead Researcher, Technology Innovation Institute 
• Juliet Shen – Product Lead, Columbia University SIPA
• Divya Siddarth* – Co-Founder & Executive DIrector, Collective Intelligence Project
• Aviya Skowron* – Head of Policy and Ethics, EleutherAI 
• Dawn Song  – Professor, Department of Electrical Engineering and Computer Science at UC Berkeley
• Joseph Spisak* – Product Director, Generative AI @Meta 
• Madhu Srikumar* – Head of AI Safety Governance, Partnership on AI
• Victor Storchan – ML Engineer 
• Mark Surman – President, Mozilla
• Audrey Tang* – Cyber Ambassador-at-Large, Taiwan
• Jen Weedon – Lecturer and Researcher, Columbia University 
• Dave Willner – Fellow, Stanford University 
• Amy Winecoff – Senior Technologist, Center for Democracy & Technology 

The post A different take on AI safety: A research agenda from the Columbia Convening on AI openness and safety appeared first on The Mozilla Blog.

Continuing our series on Anonym’s technology, this post focuses on the Transparency Portal, a critical tool designed to give our partners comprehensive visibility into the processes and algorithms that handle their data. As a reminder, Mozilla acquired Anonym over the summer of 2024, as a key pillar in its effort to raise the standards of privacy in the advertising industry. These privacy concerns are well documented, as described in the US Federal Trade Commission’s recent report. Separate from Mozilla surfaces like Firefox, which work to protect users from invasive data collection, Anonym is ad tech infrastructure that focuses on improving privacy measures for data commonly shared between advertisers and ad networks.

Anonym uses Trusted Execution Environments, which include the benefit of providing  security to users through the attestation processes. As discussed in our last post, this guarantees that only approved code can be run. Anonym wanted our customers to be able to participate in this process without the burden of overly complicated technical integration. For this reason Anonym developed the Transparency Portal and a process we call binary review. Anonym’s Transparency Portal provides comprehensive review capabilities and operational control over data processing to partners.

The Transparency Portal: Core features

The Transparency Portal is designed to offer clear, actionable insights into how data is processed while enabling partners to maintain strict control over the use of their data. The platform’s key components include:

• Knowledge Base
  Anonym provides comprehensive documentation of all aspects of our system, including:  1) the architecture and security practices for the trusted execution environment Anonym uses for data processing; 2) details on the methodology used for the application, such as our measurement solutions (Private Lift, Private Attribution) and 3) how Anonym uses differential privacy to help preserve the anonymity of individuals.
• Binary Review and Approval
  Partners can review and approve each solution Anonym offers, a process we call Binary Review. On the Your Binaries tab, partners can download source code, inspect cryptographic metadata, and approve or revoke binaries (i.e. the code behind the solutions) as needed. This ensures that only vetted and authorized code can process partner data.

• Code Comparison Tool
  For partners managing updates or changes to binaries, the portal includes a source code comparison tool. This tool provides line-by-line visibility into changes (aka ‘diffs’) between binary versions, highlighting additions, deletions, and modifications. Combined with detailed release notes, this feature enables partners to quickly assess updates and make informed decisions.

• Job History Logs
  A complete log of all data processing jobs enables tracing of all data operations. Each entry details the algorithm used, the data processed, and the associated binary version, creating an immutable audit trail for operational oversight and to help support regulatory compliance.
• Access and Role Management
  The portal allows partners to manage their internal access rights. Administrative tools enable the designation of users who can review documentation, approve binaries, and monitor processing activities.

Bridging security, transparency and control

We believe visibility and accountability are foundational requirements of any technology, and especially for systems that process consumer data, such as digital advertising. By integrating comprehensive review, approval, and audit capabilities, the Transparency Portal ensures that our partners have full visibility into how their data is used for advertising purposes while maintaining strict data security and helping to support compliance efforts.  

In our next post, we’ll delve into the role of encryption and secure data transfer in Anonym’s platform, explaining how these mechanisms work alongside the Transparency Portal and the TEE to protect sensitive data at every stage of processing.

The post Building trust through transparency: A deep dive into the Anonym Transparency Portal appeared first on The Mozilla Blog.

Giving people the ability to shape the internet and their experiences on it is at the heart of Mozilla’s manifesto. This includes empowering people to choose how they search.

On Nov. 20, the United States Department of Justice (DOJ) filed proposed remedies in the antitrust case against Google. The judgment outlines the behavioral and structural remedies proposed by the government in order to restore search engine competition.

Mozilla is a long-time champion of competition and an advocate for reforms that create a level playing field in digital markets. We recognize the DOJ’s efforts to improve search competition for U.S. consumers. It is important to understand, however, that the outcomes of this case will have impacts that go far beyond any one company or market. 

As written, the proposed remedies will force smaller and independent browsers like Firefox to fundamentally reexamine their entire operating model. By jeopardizing the revenue streams of critical browser competitors, these remedies risk unintentionally strengthening the positions of a handful of powerful players, and doing so without delivering meaningful improvements to search competition. And this isn’t just about impacting the future of one browser company — it’s about the future of the open and interoperable web. 

Firefox and search

Since the launch of Firefox 1.0 in 2004, we have shipped with a default search engine, thinking deeply about search and how to provide meaningful choice for people. This has always meant refusing any exclusivity; instead we preinstall multiple search options and we make it easy for people to change their search engine — whether setting a general default or customizing it for individual searches. 

We have always worked to provide easily accessible search alternatives alongside territory-specific options — an approach we continue today. For example, in 2005, our U.S. search options included Yahoo, eBay, Creative Commons and Amazon, alongside Google. 

Today, Firefox users in the U.S. can choose between Google, Bing, DuckDuckGo, Amazon, eBay and Wikipedia directly in the address bar. They can easily add other search engines and they can also benefit from Mozilla innovations, like Firefox Suggest.

For the past seven years, Google search has been the default in Firefox in the U.S. because it provides the best search experience for our users. We can say this because we have tried other search defaults and supported competitors in search: in 2014, we switched from Google to Yahoo in the U.S. as they sought to reinvigorate their search product. There were certainly business risks, but we felt the risk was worth it to further our mission of promoting a better internet ecosystem. However, that decision proved to be unsuccessful. 

Firefox users — who demonstrated a strong preference for having Google as the default search engine — did not find Yahoo’s product up to their expectations. When we renewed our search partnership in 2017, we did so with Google. We again made certain that the agreement was non-exclusive and allowed us to promote a range of search choices to people. 

The connection between browsers and search that existed in 2004 is just as important today. Independent browsers like Firefox remain a place where search engines can compete and users can choose freely between them. And the search revenue Firefox generates is used to advance our manifesto, through the work of the Mozilla Foundation and via our products — including Gecko, Mozilla’s browser engine. 

Browsers, browser engines and the open web

Since launching Firefox in 2004, Mozilla has pioneered groundbreaking technologies, championing open-source principles and setting critical standards in online security and privacy. We also created or contributed to many developments for the wider ecosystem, some (like Rust and Let’s Encrypt) have continued to flourish outside of Mozilla. Much of this is made possible by developing and maintaining the Gecko browser engine.  

Browser engines (not to be confused with search engines) are little-known but they are the technology powering your web browser. They determine much of the speed and functionality of browsers, including many of the privacy and security properties.  

In 2013, there were five major browser engines. In 2024, due to the great expense and expertise needed to run a browser engine, there are only three left: Apple’s WebKit, Google’s Blink and Mozilla’s Gecko — which powers Firefox. 

Apple’s WebKit primarily runs on Apple devices, leaving Google and Mozilla as the main cross-platform browser engine developers. Even Microsoft, a company with a three trillion dollar market cap, abandoned its Trident browser engine in 2019. Today, its Edge browser is built on top of Google’s Blink engine.

There are only three major browser engines left — Apple’s WebKit, Google’s Blink and Gecko from Mozilla. Apple’s WebKit mainly runs on Apple devices, making Gecko the only cross-platform challenger to Blink. Remedies in the U.S. v Google search case

So how do browser engines tie into the search litigation? A key concern centers on proposed contractual remedies put forward by the DOJ that could harm the ability of independent browsers to fund their operations. Such remedies risk inadvertently harming browser and browser engine competition without meaningfully advancing search engine competition. 

Firefox and other independent browsers represent a small proportion of U.S. search queries, but they play an outsized role in providing consumers with meaningful choices and protecting user privacy. These browsers are not just alternatives — they are critical champions of consumer interests and technological innovation.

Rather than a world where market share is moved from one trillion dollar tech company to another, we would like to see actions which will truly improve competition — and not sacrifice people’s privacy to achieve it. True change requires addressing the barriers to competition and facilitating a marketplace that promotes competition, innovation and consumer choice — in search engines, browsers, browser engines and beyond. 

We urge the court to consider remedies that achieve its goals without harming independent browsers, browser engines and ultimately without harming the web.

We’ll be sharing updates as this matter proceeds.

The post Proposed contractual remedies in United States v. Google threaten vital role of independent browsers appeared first on The Mozilla Blog.