The European Commission has ordered Android AI interoperability across eleven operating system features, ending Gemini's exclusive grip on voice invocation, screen automation and ambient context. Rival assistants get equal access on 60% of Europe's phones. Google has until Android 18 to comply.
By Blumefield | July 20, 2026
A Decision That Redraws the Mobile AI Map
On 16 July the European Commission issued two sets of binding specification measures to Google, and the more consequential of the pair mandates Android AI interoperability across eleven separate operating system features. The practical effect is blunt. Nearly every capability that currently makes Gemini feel native to an Android handset, from voice invocation to acting inside third-party apps, must be opened to competing AI services on equivalent terms.
This is not a fine. Specification proceedings under the Digital Markets Act do not determine whether a gatekeeper broke the law. They clarify what compliance actually requires, which in the case of Android AI interoperability is the more invasive instrument. A penalty is a number Google can absorb. An engineering mandate reaches into the architecture of the operating system itself.
The scale of the affected market explains why the Commission treated Android AI interoperability as urgent. Roughly 60% of European mobile users carry an Android device, and the Commission's own reasoning states plainly that Gemini, backed by Google's full-stack capabilities, is uniquely placed to become the leading AI offering on mobile. Distribution, not model quality, was the moat under examination.
What the Eleven Features Actually Unlock
The Commission's published Q&A groups the eleven features covered by the Android AI interoperability mandate into four building blocks: invocation, context, actions, and access to resources. Read individually they sound technical. Read together they describe the entire surface of a modern assistant.
Invocation covers the long-press home button and always-on hotword detection, the most visible piece of Android AI interoperability for ordinary users. Wake words can no longer be reserved for Google services, which means a rival assistant becomes summonable with the screen off or the device in standby. Context covers centralised access to on-device app data, context-aware intelligence, and ambient sensor streams from microphone, camera and screen. Actions covers structured on-device integration, screen automation and system integration, so a third-party assistant can open a delivery app, navigate it and complete an order in a background virtual window while the user does something else. Resources covers system-level on-device models, including the Gemini Nano models already shipped on Android, plus background execution rights.
The structured integration clause carries a detail that deserves attention. Google must expose operating system level integration channels for its own applications, specifically Gmail, Calendar, Drive, Docs, Maps, YouTube, Messages and Phone. A competing assistant will be able to send mail through Gmail on a user's behalf. That is a meaningful concession, and one Google did not offer voluntarily across two years of dialogue.
Google's Security Objection and Its Limits
Google's public response to the Android AI interoperability order came from global affairs head Kent Walker, who argued the decisions risk undermining privacy and security guardrails for millions of Europeans and that the rulings discount extensive evidence of user harm. The objection is not frivolous. Handing ambient microphone and screen access to arbitrary applications is genuinely a security question, not merely a competitive one.
The Commission anticipated it. The measures permit Google to impose objective and non-discriminatory eligibility conditions on the five most sensitive features, covering screen automation, structured on-device integration, system integration, centralised on-device data access and context-aware intelligence. Independent third parties certify alongside Google that an applicant meets the criteria, assessments must conclude within four weeks, and no commercial requirements may be attached. Google publishes draft terms by 1 February 2027 and final terms by 1 May 2027.
That structure is the interesting part. Brussels did not dismiss the security argument. It accepted the premise and then removed Google's discretion over the outcome, which is a template other regulators are likely to copy when they approach Android AI interoperability in their own markets. The gatekeeper keeps a veto over genuinely unsafe integrations while losing the ability to define safety in ways that happen to exclude commercial rivals. Whether four-week certification windows survive contact with a serious adversarial actor is an open question, and the Commission has reserved the right to amend as market conditions evolve.
The Search Data Decision Running Alongside
The second specification decision addresses a different bottleneck. Google must share anonymised search data, covering the query, click and ranking signals it collects to optimise its own service, with eligible third-party search providers beginning in January 2027. Crucially, AI chatbots offering search functionality are explicitly eligible recipients.
That eligibility clause quietly matters more than Android AI interoperability for some competitors. A rival assistant with system-level phone access still hallucinates if its retrieval layer is thin. The Commission's anonymisation method was developed with the European Data Protection Board in line with draft joint guidelines on the interplay between the DMA and GDPR, and the decision sets a pricing formula rather than leaving terms to bilateral negotiation, which is where previous data-sharing offers stalled.
Google retains the right to assess, before sharing, whether a specific recipient poses serious cyber security or data protection risks. As with the Android eligibility programme, the shape is the same: a preserved safety veto sitting inside a mandated process. Readers following the Brussels effect on AI markets will find this pattern familiar from the wider Blumefield coverage of European technology regulation.
Why Android 18 Is the Real Deadline
The compliance calendar is the detail most commentary has underweighted. Google must implement the bulk of the measures in Android 18, and by 1 August 2027 at the latest. Concurrent hotword detection, allowing several assistants to respond to different wake words simultaneously, slips to Android 19 and 1 August 2028.
Those dates are not generous, and they are not trivial either. Android AI interoperability of this depth requires new permission models, new certification infrastructure, documentation, beta access and technical assistance for third parties, delivered free of charge and across devices from every manufacturer in the ecosystem. Google must also report to the Commission regularly over the coming two years on design and implementation progress. The measures explicitly bar Google from pushing the engineering burden onto handset makers.
The strategic question is what OpenAI, Anthropic and others do with a two-year runway. Building an assistant that meaningfully exploits ambient context and screen automation is not a port of an existing chat application. It is a different product with a different trust posture, and it must be built for a market of roughly 450 million people under a legal framework that does not yet exist elsewhere. Some will conclude the engineering cost of Android AI interoperability is worth it. Others will wait to see whether Washington, London or Delhi follow.
What is no longer in doubt is that the assistant layer has become the contested surface. For a decade the fight over mobile was about app stores, defaults and search placement. Android AI interoperability shifts the battleground to something more fundamental, namely which software gets to see what you see, hear what you hear, and act on your behalf. Brussels has decided that question should not be answered by whoever happens to own the operating system. Google has until Android 18 to build the alternative.
Sources include the European Commission's specification decision announcement, the published measures in case DMA.100220, and Article 6(7) of the Digital Markets Act.