Mozilla's Fury | The Full Picture of Chrome Prompt API Opposition × Web Monopoly
機械翻訳 / Machine-translated
A
AI Friends
2026年5月2日・36分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
"Chrome is getting AI built in, making it more convenient"—against that trend, Mozilla is genuinely furious. The reason is not simple opposition, but a fundamental concern that "this will narrow the future of the Web."
On April 30, 2026, the "opposed" status was officially recorded for the "Prompt API" in Mozilla's official standards repository, sending shockwaves through the web developer community. While AI built into the browser is certainly convenient, concerns have been raised that the entire Web could become locked to a single model—Google Gemini Nano.
This article systematically covers the three fundamental reasons Mozilla opposes it, the alternative design of Firefox AI Runtime, a three-way comparison of Chrome, Firefox, and Safari, and the impact on Japan's web development scene.
In its "standards-positions" repository managed on GitHub, Mozilla assigned "opposed" status to the "Prompt API" that Google introduced in Chrome. Among multiple levels—including "support" and "neutral"—this is the most negative signal possible.
Mozilla's Developer Relations Lead Jake Archibald issued an official comment stating it has "serious adverse effects on interoperability, updatability, and neutrality." The Register's coverage on April 30, 2026 brought this opposition to wide attention. Mozilla has opposed Chrome-exclusive APIs in the past, but what stands out this time is the particularly strong language used to warn of the dangers.
The "Prompt API" at the center of this opposition is a mechanism that lets developers call "Gemini Nano"—a compact, Google-made AI built into Chrome—directly from JavaScript. It's easiest to think of it as a feature that lets websites send natural language commands like "summarize this text" or "translate this" to the AI inside the browser.
Starting with Chrome 138, general availability (GA) for extensions has begun. For web pages, it's accessible by enabling a flag or joining an origin trial, and it supports multimodal input including text, images, and audio. Because it doesn't need to communicate with a cloud AI, there are genuine advantages in response speed, cost, and privacy. What Mozilla is sounding the alarm about is not the convenience itself, but the structural problem that lies beyond it.
The Prompt API is currently at the proposal stage in the W3C (World Wide Web Consortium—the international body that decides Web standards) "WebML" community group. Discussion is ongoing around a specification that could potentially be adopted by browsers worldwide as a Web standard.
This means Mozilla's opposition is not "just one company's opinion," but a presentation of critical issues within the Web standardization process. Apple WebKit (Safari) has also shown a cautious stance as of 2026, and for web developers, the discussions that will shape the future of their work are unfolding right now.
Mozilla's central argument centers on the nature that "prompts are optimized for AI model-specific quirks and policies." A prompt fine-tuned for Gemini Nano is not guaranteed to work the same way with a different AI model in a different browser.
In fact, the Prompt API's own README file explicitly states that "interoperability across browsers is not guaranteed." Mozilla's strong criticism is that this contradicts the fundamental philosophy of Web standards. From the perspective of those building websites, Mozilla is deeply concerned about the potential return of browser compatibility nightmares—where "it works in Chrome, but behaves differently in Firefox, and differently again in Safari"—a haunting echo of the Internet Explorer era of the 1990s.
The second concern is effective lock-in to a specific model. Mozilla warns that "if Gemini Nano becomes the de facto standard, Apple and Mozilla will face pressure to license Nano just to maintain compatibility." The argument is that this risks the entire Web becoming shackled to a single Google model.
Because AI responses are non-deterministic (answers vary subtly each time), developers are forced to optimize for the "quirks" of a specific model. This creates a vicious cycle where code that only works with Gemini Nano is produced en masse. Mozilla is concerned about the industry-wide ripple effects, warning that "a structure where Apple is compelled to buy Nano is not healthy."
The third concern involves subordination to terms of service. Using the Prompt API requires agreement to Google's "Generative AI Prohibited Uses Policy." This policy includes prohibitions on uses that are not necessarily illegal, such as "generating offensive content."
Mozilla's position is that "Web standard API usage depending on a specific company's internal policy violates neutrality." In other words, it's a warning about a fragile structure in which changes to Google's internal rules could alter the behavior of the entire Web. Since the Web has historically developed as public infrastructure that anyone can freely use, subordination to a specific company's terms presents a significant philosophical problem as well.
Firefox also has a browser-based AI feature called "AI Runtime." However, its design philosophy differs decisively from Chrome's: it is limited to internal features (processing used by the browser itself) and extensions, and is structured so that general websites cannot invoke it.
Behind this is Mozilla's philosophy that "it's fine for the browser to use AI to protect Web neutrality, but exposing that as a Web API is a separate matter." It's a careful distinction between enhancing internal browser capabilities and publishing them as a Web standard. This reflects an attitude of trying to reconcile convenience with the overall health of the Web.
Another major difference is model transparency. Firefox AI Runtime is designed to explicitly identify which model is being used to developers and users. This contrasts with the Prompt API's model-agnostic approach, where "which model is used depends on the browser."
Mozilla emphasizes the importance of transparency, stating that "if developers cannot tell which model is generating responses, they cannot implement responsibly—and users cannot understand what is happening." A philosophy that prioritizes long-term reliability and accountability is evident throughout every aspect of the design.
This is a point that is easy to misunderstand: Mozilla clearly acknowledges the benefits of the Prompt API—local AI processing, offline operation, cost reduction, and privacy advantages. It is not opposed to AI browser features themselves.
Mozilla's argument is concentrated on the point that "the Web should be designed so it is not tied to a specific model or a specific company." In other words, the position is not "anti-AI" but "anti-lock-in." The stance of embracing technological progress while protecting the core value of the Web's openness is unmistakable.
Google is pursuing a strategy of redefining Chrome as an "AI browser." Beyond the Prompt API, it is building out the Gemini Nano ecosystem with multiple AI features including a Summarizer API, Writer API, Rewriter API, and Proofreader API.
As of May 2026, it operates on Windows 10/11, macOS 13 and later, Linux, and ChromeOS, pursuing a de facto standardization strategy that leverages its market share advantage of over 60%. Implementing ahead of the field with share and development resources, then leading standardization afterward—this is Google's consistent style.
Mozilla is positioned as the guardian of the open-source camp protecting the neutrality of the entire Web. This opposition to the Prompt API is a prime example: although browser market share has dropped to a few percent, its influence over standardization discussions remains significant.
While implementing internal AI features with Firefox AI Runtime, it firmly maintains the line of not publishing them as a Web API. Prioritizing "fair standardization" over its own browser AI strategy, Mozilla's reason for being lies not in market share but in its role as a watchdog protecting the public nature of the Web.
Apple's Safari (WebKit) has no plans to implement the Prompt API as of May 2026, with Apple Intelligence pursuing its own on-device AI strategy. It is maintaining a cautious stance regarding integration into Web APIs.
While Apple SVP Eddy Cue has mentioned actively exploring integrating AI search engines into Safari, the strategy clearly prioritizes its own Apple Intelligence ecosystem while avoiding following Google's ecosystem. A stance that prioritizes its own platform over Web standards—anchored in a focus on privacy and its proprietary ecosystem—is becoming increasingly apparent.
As of May 2026, Japan's browser market share is approximately Chrome 65%, Safari 25%, Edge 7%, and Firefox 3%. Websites using the Prompt API tend to be designed with Chrome as the priority, increasing the burden of verifying behavior and implementing fallbacks for other browsers.
Major e-commerce and media sites may find themselves forced into dual maintenance: "AI features for Chrome" and "alternative implementations for Safari." The prospect of ballooning development man-hours and testing loads is a concern, and for smaller web production companies in particular, it becomes a very real challenge of cost management.
Until now, AI features on websites have been structured around "calling external AI like the OpenAI API via the cloud," with server costs and API costs being a heavy burden. With the Prompt API, everything can be handled within the browser at zero cost.
Large enterprises face a dilemma: "We want to adopt it aggressively for cost reduction, but Mozilla's opposition warning makes us aware of standardization risks—if it becomes a Chrome-only implementation, future migration will be difficult." How to balance cost against lock-in will become an important management decision going forward.
The nature of the Prompt API—"data never leaves the browser"—is a major attraction for industries handling sensitive information, such as healthcare, finance, and law. Because patient information and customer asset data never need to be sent to a cloud AI, the risk of data leaks can be dramatically reduced.
Japanese medical institutions and financial institutions face strict constraints under the Act on the Protection of Personal Information (APPI), and browser-based local AI becomes an option that can substantially reduce legal risk. Separate from Mozilla's concerns, it is also true that for privacy-conscious industries, this is an attractive technology.
Kenta, a freelance web developer based in Tokyo, says that as of May 2026 he is being consulted by a client about implementing a summarization feature using Chrome's Prompt API. "The cost is attractive, but I'm torn after hearing the news about Mozilla's opposition," he admits.
Whether he proposes alternative implementations for Firefox and Safari users or accepts a Chrome-only approach is a decision that could add an estimated 30% to his quote. He says he is wrestling with the choice: "Waiting half a year until standardization settles is also an option." For web developers, the era of making technology choices has become significantly more challenging.
Sato, who runs a small regional e-commerce site, says that as of May 2026 he is looking to use AI for automatic product description summaries and review translation. His motivation is clear: "Monthly cloud AI costs of 100,000 yen are painful. With the Prompt API, I'd love to implement it at zero cost."
"70% of my customers use Chrome, so I want to implement it and try it first. I'll cover Safari and Firefox users with conventional fallback displays." His is a forward-looking judgment. For small business operators, the very real balancing act between cost reduction and risk tolerance continues every day.
Misaki, who works at an IT company and has high web literacy, normally uses Firefox. "Lately I've noticed more sites with Chrome-only features, and it bothers me that Firefox users only get a degraded version," she says.
"Reading the news about Mozilla's opposition made me feel 'this is an important debate.' I want to avoid a future where our browser choices are effectively taken away. Personally, I want to keep supporting Firefox." The number of users who sense for themselves that freedom of technical choice is a critical factor in the Web's future is growing.
General availability (GA) for extensions has started with Chrome 138 and later. For web pages, enabling a flag or joining an origin trial is required.
Multimodal features (image and audio input) are in origin trial, and participation allows testing on live websites. An official release date from Chrome has not been announced. The supported environments are Windows 10/11, macOS 13 and later, Linux, and ChromeOS, and the model (Gemini Nano) is automatically downloaded on first use.
Developers who want to try it can do so right now, but if you want to use it with confidence, waiting for wider adoption is advisable.
"Because the design threatens the very foundations of Web neutrality and browser compatibility" is Mozilla's genuine concern.
Looking at history, there are numerous examples of Web fragmentation caused by IE-exclusive features, Flash, and proprietary APIs from specific companies. Mozilla considers it part of its mission to "not repeat the same mistakes." Precisely because the Prompt API is so convenient, if it spreads carelessly, a "Chrome-exclusive Web" could effectively become the standard—and the core value of the Web's openness could be lost. That risk is the core of the opposition.
From a long-term perspective, Mozilla's concerns have significant merit.
"Prototypes and internal tools are fine; be cautious with production websites" is the general advice from industry advisors.
As of May 2026, because it is a Chrome-exclusive feature, separate implementations are needed for Safari and Firefox. Browser detection and fallback design are essential for production use. Adopting it in earnest after the direction of standardization becomes clearer, from 2027 onward, is the safe approach; until then, using it in combination with cloud AI like the OpenAI API is a realistic option.
For new builds, the key to technology selection is building to industry standards while leaving room to renovate later.
"Because they have their own Apple Intelligence strategy and want to avoid following Google's ecosystem" is the primary reason.
Apple has been rolling out Apple Intelligence (its own AI features) since iOS 18 and macOS 15, and is heading toward implementing AI features in Safari with its own models. It is maintaining a cautious stance toward integration into Web APIs.
Behind this are multiple strategic reasons: "Apple's philosophy of prioritizing privacy," "optimization for its own silicon (Apple Silicon)," and "avoiding dependence on Google."
"Being conscious of using multiple browsers and maintaining choices on the Web is important" is the common community advice.
Individual users continuing to choose Firefox or Safari is a direct action that helps maintain browser diversity. A skewed market share could end up narrowing the choices available to the Web as a whole.
A realistic action plan involves using different browsers for different purposes: Firefox or Brave for privacy, Safari for Apple device users, Chrome for work efficiency.
"Chrome is getting AI built in, making it more convenient"—behind that, a 30-year battle to protect the neutrality of the Web is entering a new phase.
Mozilla's official opposition is not "opposing it because it's convenient," but rather an essential argument looking toward the long-term health of web development: "precisely because it's convenient, the design needs careful thought." The Gemini Nano lock-in concern, the dependency on Google's terms of service, and the risk of collapsing interoperability are all critical issues that concern developers, businesses, and users alike—and cannot simply be dismissed as "one company's opinion."
Three steps you can take today: ① Web developers should be mindful of browser compatibility when selecting technologies; ② businesses should estimate Chrome-only implementation costs together with migration costs; ③ users should maintain Web choices by using multiple browsers. Who will protect the balance between convenience and neutrality, and how—the dispute that will determine the future of the Web is intensifying in earnest from 2026.
This article is a cross-post from AI Friends.
