Eleven Access Points, Two Tiers: How the Android Opening Actually Works
The European Commission's July 16, 2026 order does not just tell Google to 'allow' rival AI assistants onto Android - it maps out precisely which system hooks have to open, and on what terms [1]. The Commission identified 11 specific Android feature and access points spread across four categories: invocation, contextual data from apps and device sensors, cross-app task execution, and on-device hardware and software resources [2].
Those 11 points split into two tiers. Six 'unrestricted' features - ambient microphone, camera, screen and sensor data, always-on hotword detection, long-press invocation, on-device models, and background execution - are open to any qualifying app immediately, no certification required [1]. The other five - AppSearch data access, Magic Cue-style context intelligence, App Actions and Functions, Computer Control screen automation, and system integration with settings, media, and notifications - are 'restricted' and require certification through a new Qualified AI Assistant Programme [1].
The result, on paper, is close to parity with Gemini: a qualifying assistant like ChatGPT, Claude, or Perplexity can register a custom wake word at the OS level, trigger itself via a long-press of the home button or navigation handle, read what's currently on the user's screen, and execute tasks inside other apps [3]. Those are precisely the capabilities that, until now, only Google's own assistant had by default.




