Mozilla Just Drew a Line Against Chrome's AI-in-the-Browser Push

Imagine opening Chrome and calling an AI model with a single line of JavaScript — no API key, no server, no bill from OpenAI. That’s the pitch behind Google’s Prompt API. Mozilla just said no, loudly, and the objection isn’t competitive sour grapes. It’s a structural critique of what the web is supposed to be.

What Prompt API Actually Does

Prompt API exposes Chrome’s built-in Gemini Nano model to any webpage through a JavaScript interface. For developers, the appeal is obvious. Summarization, translation, classification — all running locally in the user’s browser, no inference bill, no rate limits, no Anthropic or OpenAI middleman.

Google wants this in the Web Platform itself. Not a Chrome extension, not an opt-in flag — a standard alongside fetch and localStorage, something every browser would be expected to ship.

Problem One: The Web Was Built on Determinism

A foundational rule of web standards is that the same input produces the same output. Math.sqrt(4) returns 2 in Chrome, Safari, Firefox, and a 2015 build of Edge. Always.

LLMs don’t work that way. The same prompt yields different answers across runs, and a quiet model version bump can rewrite behavior overnight. Mozilla’s argument: standardize this and you get a web where identical code behaves differently depending on which browser, which version, which Tuesday. That’s not a bug to paper over. It’s a direct hit on the trust model that’s held the web together for 25 years.

Problem Two: Whose Model Wins

If Chrome ships Gemini Nano, Safari will reach for an Apple model, Firefox for something else. Same site, same code, three different answers depending on which app the user happened to open.

The deeper issue is bias and hallucination baked into the platform. There’s no neutral way to decide which political opinions a model expresses, which facts it gets wrong, which topics it refuses. Mozilla’s framing is sharp: browser vendors become de facto content gatekeepers. The search-algorithm fights of the last decade were a warm-up. This pushes the same battle one layer deeper, into the runtime itself.

Problem Three: Standards by Fait Accompli

Web standards normally go through W3C or WHATWG, where multiple vendors hash things out before anything ships. It’s slow on purpose — slow is what stops one company from reshaping the web in its image.

Google’s playbook here is different: ship Prompt API in Chrome first, let developers build on it, then standardize after the fact. Mozilla sees the pattern and names it — the embed tag and AMP, all over again. With Chrome holding north of 65% of desktop market share, “ship and standardize later” isn’t a proposal. It’s a fait accompli wearing a process costume.

The Real Question

Browser-integrated AI isn’t inherently bad. Explicit user opt-ins, extensions, OS-level assistants — there are plenty of designs that don’t break the web. The specific thing Mozilla is fighting is making it a standard API any website can invoke without the user knowing.

This isn’t a Mozilla-versus-Google story. It’s a question about what the web is for. Are you ready for a future where your site’s behavior depends on whichever model Google felt like shipping last quarter? Mozilla isn’t, and they’re betting the rest of the platform shouldn’t be either.