EU’s Key Order Forces Android Access for Rival AI Assistants

• Android AI assistants may receive access to sensors, screens and background controls that Google currently reserves largely for Gemini.
• The EU’s Android AI assistants ruling requires Google to deliver key interoperability changes with Android 18 or by August 2027.
• Certified rival assistants could draft Gmail messages, manage Calendar events, control media and operate compatible apps through screen automation.
• Google can impose security certification for sensitive capabilities, but the Commission sharply limits how it can deny or suspend access.

Google’s assistant advantage is now the target

Android AI assistants are about to become a much bigger battleground in Europe. The European Commission has told Google that it cannot keep the deepest parts of Android largely reserved for Gemini while treating rival assistants as ordinary apps knocking at the door.

The order, adopted on July 16 under the EU’s Digital Markets Act, is unusually specific. Google must provide qualifying competitors access to hardware and system abilities that make an assistant feel genuinely built into a phone: camera and microphone feeds, what is visible on screen, low-power wake-word detection, background execution, and tools that can operate other apps. For Android AI assistants, the deadline is Android 18, or August 1, 2027 at the latest.

At first glance, this looks like dry interoperability policy. It goes to the central question of the AI phone era: does your chosen assistant live on the device, or does it sit in a browser-like sandbox waiting for you to manually hand it information?

Google has spent years building the former experience around its own services. Gemini can understand what is on your display, tie requests into Android’s system layer, and reach across Google’s sprawling app portfolio. Rival Android AI assistants, by comparison, have often had to work like a guest who can enter the house but not touch the thermostat.

My read is that Brussels has identified a real problem before it becomes permanent. Once people build habits around an always-available assistant, the default choice becomes sticky in the same way default search once did. And Google knows more than most about the value of defaults.

Android AI assistants would get the tools that matter

The Commission’s specification covers 11 Android features, splitting them into two camps. Five can sit behind a certification process, while six must be open to third parties without Google deciding which kinds of assistant deserve access.

For Android AI assistants, that open group is the headline act. It includes ambient data from the microphone, system audio, camera, display, location and device sensors; always-on hotword detection; long-press invocation; access to system-level on-device models; third-party model implementation; and background execution. Consent still applies, as it should. But the Commission says rivals should receive the same consent path and privacy indicators available to Google’s own services, rather than being forced through more burdensome permission prompts every time they need context.

For users, this is the difference between saying “ask my assistant about what I’m looking at” and having to take a screenshot, save it, open another app, upload the file, then explain the question again. Nobody wants that workflow. It is the digital equivalent of writing a shopping list on paper before asking a store clerk whether they sell milk.

There is an important hardware catch. Wake-word detection relies on a phone’s low-power digital signal processor, so it will only work on handsets that have the necessary chip support. Google also gets more time to support several listening assistants at once: that requirement lands with Android 19, no later than August 1, 2028. Anyone imagining five bots silently competing to hear their name can relax, at least for a while.

Certification is allowed, but Google cannot turn it into a moat

The other five features are more sensitive, and the EU has given Google room to demand certification. They include access to app-shared on-device data, context-aware intelligence, structured app actions, screen automation, and broader system integration such as notifications, settings, media controls and screenshots.

Those categories unlock the practical stuff. A certified service could retrieve or draft Gmail, create Calendar events, pull material from Drive and Docs, start Maps navigation, manage YouTube playback, access Messages, and place calls. In other words, Android AI assistants could move from answering questions to completing the small, annoying tasks that fill a phone user’s day.

Google must create a Qualified AI Assistant Programme and accept certifications issued by approved independent Trusted Certification Authorities. The certification itself must be free for developers. For Android AI assistants seeking sensitive capabilities, Google can test whether an assistant confirms user intent before irreversible actions, limits accidental disclosure, meets baseline mobile security expectations, and resists agentic behavior that could override what a user actually meant.

Frankly, those are sensible safeguards. An assistant that can send a message, make a call, or shuffle through a banking app needs a higher standard than a weather widget. The key constraint is that Google cannot quietly add extra hurdles that Gemini does not face. The Commission must approve any demands that go beyond the defined security boundaries.

Google’s ability to suspend an assistant is narrow too. It needs consistent evidence of practices causing severe and immediate harm, must share that evidence with regulators and certification authorities, and faces a one-month appeal timetable. That is a deliberate attempt to prevent security from becoming a convenient label for exclusion.

App makers should pay attention before Android 18 arrives

There is a less obvious consequence here for developers. By the deadline, a certified assistant — or an uncertified one a user has explicitly permitted — may be able to open an app on a virtual display, inspect its screen and tap through its interface in the background. Android AI assistants could make travel booking, expense filing, shopping and support workflows much less tedious. It could also expose plenty of brittle app design.

The Commission’s rules allow controlled apps to shield sensitive views from the controlling assistant. Developers should use that option for payment details, medical records, authentication flows and anything else that does not belong in an autonomous workflow. Google may offer broader controls to block automation in parts of an app or stop an app’s context reaching proactive suggestion systems, but it is not required to do so.

That distinction matters. If you build an Android app, waiting until the Android 18 beta to think about agent access is a poor bet. Map your sensitive screens now, decide where explicit confirmation belongs, and assume a user will eventually ask an assistant to complete a task you designed around manual taps.

The search-data order may matter almost as much

The Commission issued a second specification decision at the same time, requiring Google to provide anonymized Search query, click and ranking data to competing search engines and AI chatbots that offer search. Access can carry a cost-based fee, but Google cannot simply sit on the behavioral data that helps its own search products improve.

Data sharing of this sort is politically explosive for obvious privacy reasons, so the published approach calls for multiple anonymization steps that remove direct identifiers and attributes that can reconnect records. The practical test will be whether the dataset remains useful after those safeguards. A rival cannot build a credible search product from a bag of statistical confetti.

Both decisions sit within the DMA’s broader effort to make designated gatekeepers open core platforms to competition. The law’s official framework is available through the European Commission’s Digital Markets Act portal. These are not fines, and they do not mean Google has already been found non-compliant. They tell the company what compliance must look like. A separate enforcement case, with penalties, remains available if Google fails to deliver.

Android AI assistants now face their real test

The Commission can force doors open. It cannot make anyone build an assistant people trust with their inbox, their camera and the ability to press buttons on their behalf. That part is on OpenAI, Anthropic, Microsoft, Perplexity, European startups and, yes, Google’s Android partners.

Still, the ruling removes a powerful excuse. If Android AI assistants fail to challenge Gemini after they can access comparable system plumbing, the problem will be product quality, privacy, pricing or user trust — not that Google kept the keys to Android’s most valuable rooms. That is a cleaner contest than the mobile industry usually offers, and it is one worth watching closely.

Frequently Asked Questions