Daring Fireball: European Commission: ‘Guidance to Google for AI Interoperability on Android & Sharing of Google Search’
By John Gruber
Archive
The Talk Show
Dithering
Projects
Contact
Colophon
Feeds / Social
--><br>Sponsorship
WorkOS MCP: Manage your auth platform from any AI agent.
European Commission: ‘Guidance to Google for AI Interoperability on Android & Sharing of Google Search’
Tuesday, 21 July 2026
The European Commission, last week:
Today, the European Commission has issued two sets of binding<br>specification measures to Google under the Digital Markets Act.
The aim of the first specification measures is to ensure that<br>competitors’ Artificial Intelligence (AI) services can compete<br>with Google’s own AI services, such as Gemini, by having equal<br>access to features on Google’s Android devices.
The aim of the second specification measures is to rebalance the<br>playing field by giving third-party search engines access to<br>search data that only Google Search can collect at scale.
They provide separate “Q&A” overviews of the guidance for Android AI interoperability and web search sharing, and the full guidance documents are PDFs (Case DMA.100220 for Android AI, Case DMA.100209 for web search). I suggest reading the two Q&A overviews, unless you’re having trouble falling asleep at night, in which case you’ll love the full PDF decisions.
Both decisions are interesting. With search, Google is required to share with competitors — search engines and AI chatbots alike — a massive amount of user data from Google Search user interactions. What terms people search for, what they click on in results, what languages and devices they use. It’s all ostensibly anonymized but that’s tricky when it comes to search terms. A lot of the terms people type into web search fields are to some degree personally identifying. The EC seems to be saying it’s Google’s problem to filter out things like passwords and usernames and omit them from the shared datasets. Google can charge money for this access, but only under “fair, reasonable, and non-discriminatory (FRAND)” prices, based on a Commission-defined methodology.
More interesting to me, however, is the guidance pertaining to on-device AI on Android devices. What the EC is dictating to Google is just breathtaking in scope. The EC is demanding that Google create APIs that allow third-party AI assistants to do everything Google Gemini does now, including:
Control hardware buttons (to invoke the assistant).
Capture anything on screen, from any app.
Seemingly unfettered access to microphones and cameras and other sensors on the device.
Unfettered background operation. Third-party AI assistants must be permitted to execute in the background whenever they want, for as long as they want.
Execute their own audio models on the digital signal processor, so they can listen at all times for their own custom “Hey Dingus” wake phrases/hot words.
Google must allow concurrent access to always-on hot word detection. So if you have Claude and ChatGPT and Grok and Meta AI installed, all of them — in addition to Gemini — must be permitted to have always-on audio detection concurrently. Google is permitted to do some vetting here, but this seems like madness.
Third-party models get access to Google’s on-device local models.
Also, in my reading, the EC is demanding that Google make available to third-party AI assistants all information in Google’s own apps (Gmail, Google Calendar, Google Docs, Google Maps, etc.) that Gemini has access to. There is no opt-out for Google regarding data from their own apps. Nor, I think, does this guidance allow third-party apps from other developers to only support specific system-level AI models. Like, let’s say you’re Slack, and you use the APIs to make the content from within Slack available to Gemini on the device. These guidelines don’t permit Google to allow Slack to say that they trust Gemini but only Gemini. If a third-party app like Slack supports making its data available to any system-level AI provider, it must make its data available to every system-level AI provider.
There’s a lot more. Basically, though, the EC is demanding that third-party AI assistants be enabled to become part of the system software, not just apps. I’m sure some people think this is a great idea. It’s the user’s device, they should be allowed to make ChatGPT or Claude or Meta AI part of their OS if they want. It’s up to them. Put users in control.
This is how PCs have traditionally worked, but many normal people’s PCs are a mess of third-party software running in the background. That includes the Mac. If you ask a normal person “What third-party software runs in the background on your Mac or PC?” they would have no idea. It’s all just magic to them. If Google supports this guidance, it could turn Android phones in the EU into PCs. Honestly, some of the stuff the EC is requiring is lower-level than what MacOS and Windows allow third-party software to...