Mozilla warns: Google is about to lock down your browserʼs AI and there will be nothing you can do about it

The next wave of AI in the browser is arriving fast, and some of the people who helped build the open web are sounding the alarm. Mozilla argues that the future is being shaped behind closed doors, with key decisions landing in one company’s hands. For everyday users, that could quietly limit what you can run, what you can choose, and what you can even know is happening under the hood of your tab.

How a closed AI browser could look

Picture your default browser quietly replacing core tasks with a single assistant that sits everywhere you type. The AI is “on-device,” but the only on-device model available is the one that comes from a single vendor, and the privileged APIs that make it useful only work in that vendor’s engine.

Small changes add up: a translation that only ships if you’re signed into the right account, a summarizer that refuses to run for other browsers, and AI prompts that show up because they’re hardwired into the UI rather than added by a permissioned extension. “When the assistant is the browser, choosing a different AI stops being a real choice,” warns one critic of lock-in.

The playbook is familiar

If this sounds abstract, think back to earlier pivots. Ad tech experiments like FLoC tried to reshape tracking, Chrome’s Manifest V3 rewrote the rules for extensions, and AMP bent the web toward a proprietary format served from a single cache. Each time, the most powerful browser could “propose” a path and then ship it at scale.

AI raises the stakes because the runtime—the place where models execute, cache, and “see” your data—lives directly inside your tabs. The entity that controls the runtime controls the defaults, the performance envelope, and the interfaces that outside tools must ask permission to use.

What that means for you

A locked AI layer can feel “helpful” while gradually shrinking choice. You’ll see fewer knobs to pick your model, fewer ways to connect your own data, and more pressure to accept one company’s policies about what counts as safe or allowed.

Your privacy surface grows too, because on-device AI is still fed with your local context—open pages, copied text, saved files—and can upload summaries or signals back to the cloud. “On-device” is not the same as offline, and controls can be deliberately thin.

Finally, true competition gets harder when the best interfaces or fastest pipelines are only available to the incumbent. Third-party tools become second-class citizens, and innovation moves at the pace of a single roadmap.

Signals to watch for

• Exclusive “system” AI features that do not expose equal APIs to other browsers or engines.
• AI permissions that are bundled into the sign-in or sync flow, rather than clear, per-feature prompts.
• Performance wins gated behind private interfaces or branded hardware-only paths.
• Language in policies that forbids competing assistants from integrating at the same depth.

Standards can fix this—if we insist

The open web has a way to keep vendors honest: transparent standards built in public, adopted by multiple engines, and tested by independent developers. For AI, that means defining neutral APIs for local model execution, safe capability prompts, and predictable permissions.

Think of a “model slot” approach: the browser exposes a common runtime, and any audited model can plug in—local or remote—under the same rules. You choose the provider, and the browser enforces the guardrails. “AI should feel like part of the web, not a store you have to enter,” goes a refrain among open-standards advocates.

Regulators are watching the defaults

In places like the EU, new laws already limit self-preferencing and lock-in through mandatory interoperability and choice screens. If an AI assistant becomes a de facto gatekeeper, expect closer looks at ranking, access to private APIs, and the treatment of rivals in core workflows.

Transparency will matter too: if a browser’s AI rewrites your page, inserts summaries, or annotates links, you should see clear labels, clear opt-outs, and clear logs. “Invisible AI is the most dangerous kind,” privacy groups point out.

What Mozilla wants to see

The open-source camp is pushing for an AI layer with vendor-neutral hooks, parity access for extensions, and a simple way to swap in different models—local, cloud, or hybrid—without losing key features. They want permissions that are front-and-center, not buried in sync settings or account ties.

Equally important is performance fairness. If there’s a fast path—say, hardware acceleration or WebGPU—it must be documented, testable, and equally available to other engines. No hidden switches, no “works best with our model” footnotes.

What you can do right now

Start by choosing a browser that treats AI as a replaceable component, not a fixed master switch. Look for clear disclosure of where your data goes, what runs on-device, and what gets uploaded for processing.

Try multiple assistants, and ask whether you can set them as true defaults across the browser’s surfaces—context menus, omnibox, page actions, and compose fields. If you can’t swap them, it’s not real choice.

Pay attention to developer signals too: when extension authors say their AI tools can’t access the right APIs, that’s often a canary for creeping lock-in. The healthier the third-party ecosystem, the healthier your long-term options.

The web’s next fork

The web survived past attempts to centralize power because users, developers, and regulators pushed back with standards and real alternatives. AI will test those defenses again, and the clock is already ticking.

As one open-web advocate put it, “Your browser should be your toolbox, not someone else’s storefront.” If we want AI that is transparent, interchangeable, and genuinely useful, we have to demand it—before the defaults quietly make that impossible.