This feels like one of those things where "just let users pick their AI assistant" sounds simple until you realize that means access to screen contents, local data, app control, wake words, etc.

At that point, who’s actually on the hook when some third-party assistant screws up: Google, the AI company, or the EU for requiring the access in the first place? Also, wouldn’t Google’s easiest move just be to pull the Gemini integrations in the EU?

Well, exactly. Just like Apple is declining to bring "Siri AI" to the EU because of this.

Interoperability rules make sense and benefit consumers when applied to many things. This isn't one of them, and the EC would do well to apply some human judgement here.

For any fully-integrated AI assistant,

a) I already more-or-less implicitly trust the OS vendor with the data the OS has access to. Allowing arbitrary vendors to offer such an assistant would mean adding an arbitrary number of other companies to that list of, currently, one.

b) The OS vendor needs to put in a huge amount of additional effort to make a public API capable of replacing their own assistant.

Or. Y'know. Just not offer it in the EU. And what are consumers going to do in response? Switch to Android, which....also isn't going to offer it?