Mozilla's Firefox team has publicly opposed Chrome's proposed LLM Prompt API, a standard that would let websites tap into language models directly through the browser. The API, documented in the webmachinelearning/prompt-api GitHub repo, is framed as meeting user expectations that browsers should have built-in LLM access. Mozilla isn't buying it.
Browser vendors are split. Google wants to ship a high-level API that abstracts away model details. Apple has pushed for lower-level primitives like the Web Neural Network API instead, citing concerns about "model fingerprinting" that could track users across sites based on unique model behaviors. This stance echoes Apple's preference for on-device processing. Microsoft, running Edge on Chromium and busy stuffing Copilot into everything, has stayed quiet on the specific API proposal.
The practical questions are hard. How do users pick between models? What hardware does this require? Typical laptops weren't built for local inference. Some developers suggested browsers agree on standardized "web-safe" models that work identically everywhere, like how CSS handles fonts. That's cleaner than what Chrome is proposing with its own on-device Prompt API.
Mozilla's opposition matters because it signals a philosophical fight about what browsers should become. Google wants the browser to be an AI platform. Mozilla wants to slow that down, citing privacy and the open web. Given Google's dominance in both search and browser market share, that tension isn't going away.