AI distribution is moving from the app store into the operating system.
That is the real signal in the European Commission’s latest Android move. This is not just another Google competition case with a new acronym attached. It is a fight over whether rival AI assistants can live inside the phone with the same practical access as Gemini.
On April 27, the Commission said it had sent Google preliminary findings under the Digital Markets Act, setting out draft measures for Android interoperability. The measures are meant to ensure that third parties have effective access to key Android capabilities, especially where competing AI services need to interact with apps and execute tasks on behalf of users. The Commission said Google currently largely reserves these capabilities for its own AI offerings on Android phones and tablets.
That is the market structure question hiding inside the phone.
The old mobile-platform fights were about app distribution, default search, payments and browser choice. Those still matter. But AI assistants change the control point. The assistant is not just another app icon. It wants to wake from anywhere, read context, act across apps, change settings, keep running in the background and make itself feel like part of the device.
If only the platform owner’s assistant can do that well, choice becomes decorative.
The AI Assistant Is Becoming A System Layer
The Commission opened the Android specification proceedings on January 27 under DMA Article 6(7), which requires gatekeepers to provide third parties with free and effective interoperability with hardware and software features controlled through an operating system. The January notice said the proceeding focused on features used by Google’s own AI services, such as Gemini.
The April case summary makes the shift explicit. It says the smart mobile device market is at a technological inflection point and that AI services are becoming central access points on mobile devices. It also says about 60% of mobile users in Europe own a Google Android smart mobile device, making Android a crucial gateway for standalone AI services to reach end users.
The gateway language matters. A phone OS controls more than installation. It controls presence.
A rival assistant can be available in the app store and still be structurally weaker. If it cannot be invoked naturally, cannot see enough context, cannot execute actions across apps, cannot access the same device resources, and cannot survive in the background, it becomes a chatbot in a box. Gemini becomes the ambient layer.
That is why the proposed measures are so specific. The Commission is not asking Google to post a fairness statement. It is asking whether the rival assistant gets the machinery required to behave like a real assistant.
Wake Words Are Distribution
One draft measure would let competing AI services be easily activated by users with a custom wake word. That sounds minor until you remember how assistants are used.
Voice activation is not branding. It is distribution. If one service responds naturally from the device while another requires opening an app, the first service gets more queries, more habit formation, more context and more chances to become the user’s default interface.
The case summary goes beyond wake words. It covers contextual invocation, access to user context and data, integration with other apps, execution of actions, system services and access to device resources. It describes the features a provider needs to offer comparable responsiveness, visibility, functionality and utility to the end user.
That is the right unit of analysis. AI competition is not only model quality. It is where the model is allowed to sit.
The Commission’s examples are mundane on purpose: sending email, ordering food, sharing photos, scheduling events, finding photos, answering questions about YouTube videos, helping with navigation in Maps, changing Bluetooth, pausing music. The boring examples are the point. Operating-system power is made of boring permissions.
This Is A Default Fight, Not A Model Fight
The strongest version of Google’s position will likely be security and privacy. That is not a fake concern.
An assistant with deeper access can do more damage. It can mishandle personal context, act in the wrong app, execute the wrong step, leak data, or turn a vague user request into a mess with system privileges. Opening OS hooks to third parties is not like letting a weather widget resize itself.
The Commission knows this. The case summary says Alphabet may take strictly necessary and proportionate measures to ensure interoperability does not compromise the integrity of the operating system, hardware or software features, within defined timelines. That is the hard part. The EU wants rivals to get meaningful access without turning Android into a permissions landfill.
The policy tradeoff is real. Too little access, and Gemini becomes the only assistant that feels native. Too much unmanaged access, and users get security theater with better voice synthesis.
But the default problem cannot be dodged. If the platform owner can cite safety to keep rival assistants away from the most useful hooks while its own assistant uses them, the competitive outcome is predictable. The firm with the OS wins the AI interface before the model comparison starts.
That is why this case reaches beyond Google. Apple, Samsung, Microsoft and every major platform company are watching the same line form. AI assistants are becoming interfaces to the operating environment. Whoever controls the environment can decide which assistant feels competent.
The Deadline Creates The Shape
The process is still preliminary. The Commission invited third parties to comment by May 13. Its case summary says it expects to adopt final binding measures by July 27, after feedback and further investigative steps.
That timetable is fast in regulatory terms and slow in AI-product terms. Product teams can ship several assistant updates before the final decision lands. The gap matters because mobile AI habits are forming now. Defaults set early can harden before remedies arrive.
For developers of rival AI services, the draft measures are a map of what they should ask for: invocation, context, app actions, first-party app integration, system integration, on-device model access and background execution. That is the stack required to compete with a native assistant.
For Google, the same map becomes a compliance architecture. It will need to explain which hooks can be opened, under what consent model, with what security boundaries, and on what timeline. “Trust us” will not be enough. “Trust everyone” would be worse.
The Implication
The Android DMA case shows where AI competition policy is going.
Regulators are moving past model access and into interface control. They are asking who can be summoned, who can see context, who can act, who can persist, and who can become the user’s daily intermediary. That is the distribution layer for agentic AI.
If Europe gets this right, Android users may be able to choose an assistant that is not structurally kneecapped at the OS boundary. If it gets this wrong, it will either entrench the platform assistant or create a risky access regime that users cannot reason about.
The narrow legal question is whether Google’s Android complies with DMA interoperability obligations.
The larger product question is simpler: when AI becomes the way people operate the phone, is the assistant market still open if only the OS owner gets the real controls?
Europe has decided that question is no longer theoretical. Good. The phone is where AI defaults become habits. Habits become markets.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.