The European Commission has ordered Google to give rival AI assistants substantially deeper access to Android—access now reserved largely for Gemini—and to share anonymized Google Search data with eligible competitors beginning in January 2027. The binding Digital Markets Act specifications, adopted on July 16, give Android’s AI-agent race a concrete regulatory timetable rather than another broad antitrust warning, as first reported by Computerworld and detailed by the Commission.
For Android users in the EU, the eventual change could be as visible as choosing a different assistant and waking it by voice while the screen is off. For developers, it is more consequential: the Commission says competing services must be able to invoke themselves from system entry points, read user-approved context, take actions inside apps, run tasks in the background, and access appropriate on-device AI resources on terms comparable to Google’s own services.
Google has sharply criticized the decisions. Kent Walker, Google’s president of global affairs, said the requirements could weaken privacy and security protections that currently govern powerful system access and the sharing of search data. The Commission, however, says its decision includes consent, integrity, privacy, and cybersecurity safeguards—and that Google can assess serious risks before releasing data to a particular recipient.
The critical detail is that this is not a fine. It is a specification decision: Brussels has now spelled out what DMA compliance is expected to look like. Failing to build and ship the required interoperability mechanisms would open the door to a separate enforcement fight.
The Commission’s Android decision covers 11 operating-system capabilities which it describes as the building blocks for capable AI services. They include invocation methods, contextual data access, actions across apps and the operating system, on-device models, and background execution.
That reaches far past the familiar ability to set a third-party app as a default assistant. Google will have to ensure that a rival service can be activated from the long-press home control or an equivalent navigation gesture, and through an always-listening hotword mechanism. The Commission explicitly says users should be able to call their preferred assistant even when the display is off, the device is idle, or battery saver is enabled.
The order also targets the capabilities that turn a chatbot into an agent. Under the framework described by the Commission, a user-approved assistant could interact with apps to draft and send an email, add a calendar event, create a note, order groceries, or book transportation. It could use structured integrations where apps expose tasks through Android’s App Functions, or complete multi-step workflows through “Computer Control,” an automation mechanism that mimics user interaction in a separate virtual window.
That is the technical territory where Gemini has enjoyed a major distribution advantage. An AI app can be excellent at conversation yet remain a relatively shallow mobile utility if it cannot see relevant on-device context, invoke itself naturally, or execute a task without forcing the user through each app manually. The Commission’s central argument is that Google cannot keep those system-level pathways for Gemini while presenting Android as open to rival AI providers.
The decision also covers access to microphone, camera, screen, speaker, sensor, and location signals under the same consent and awareness conditions applied to Google. It calls for comparable access to on-device AI models—including Gemini Nano models designated as part of the Android operating system—and comparable hardware and background-execution conditions for third-party local models.
That last point matters beyond the headline fight over Gemini. Local models can offer lower latency, offline functions, and less cloud exposure for tasks such as transcription, proofreading, and summarization. If implementation follows the Commission’s direction, an assistant from OpenAI, Anthropic, Microsoft, Perplexity, Mistral, or a smaller European provider would have a clearer route to building richer Android experiences without having to reproduce every underlying system capability itself.
That is an important distinction for enterprise Android deployments. A corporate mobility team may welcome more capable assistants for note-taking, translation, scheduling, and accessibility, but it will not want an unvetted consumer AI app able to process screen contents or interact with business apps. The operational question will be whether Android, OEM tools, enterprise mobility management platforms, and the rival assistants themselves can present permissions in a way that is intelligible, revocable, auditable, and compatible with managed-device policy.
Google’s objection is therefore not trivial. The company argues that its existing controls include vetting third-party assistants and that forcing access to sensitive system interfaces can expand the attack surface. A poorly designed integration could create fresh routes for data leakage, overbroad automation, malicious prompt injection, or abuse of accessibility-like controls.
But the Commission rejects the idea that security requires Google-exclusive access. Its position is that parity can be delivered with safeguards and that the DMA requires effective interoperability, not a permission model so restrictive that rival products cannot compete. The ruling gives Google room to assess whether a particular data recipient raises serious cybersecurity or data-protection risks, while also making clear that broad safety claims cannot be used as a permanent barrier.
For security teams, the practical result is likely to be more scrutiny of AI-assistant permissions, not less. The arrival of agents with the ability to observe context and act across apps will make Android’s permission prompts, managed configurations, app allowlists, and logging increasingly central controls.
The Commission says Google’s previous proposal was ineffective because it eliminated between 90% and 100% of unique queries from the data set and excluded AI chatbots with search features from the pool of recipients. In Brussels’ view, that left prospective competitors unable to obtain a useful signal for improving ranking, retrieval, query interpretation, and other core search functions.
Google Search’s advantage is not merely its interface or index. It is the enormous feedback loop built from queries, clicks, refinements, and the resulting quality improvements. A new search product or AI answer engine can build a strong model and still struggle to match the practical relevance of a system trained by years of web-scale behavior.
The Commission is trying to narrow that gap without releasing identifiable search histories. Its specifications require a multilayer anonymization approach developed with privacy expertise and aligned with draft guidance from the European Commission and European Data Protection Board. The data may only be used to develop and optimize search services, rather than for unrelated advertising, profiling, or model-training purposes.
Recipients will not receive the data for free. The Commission’s decision allows Google to recover the incremental cost of providing access under a defined pricing framework. That choice is designed to avoid turning the decision into a forced transfer of a valuable asset while preventing price or process from becoming another form of exclusion.
Those dates leave Google, Android device makers, app developers, AI providers, and regulators with a difficult implementation period. The Commission has not dictated every API design, permission screen, certification workflow, or background-processing limit. Google retains engineering discretion, but not discretion to reserve the practical benefits of a newly covered capability for Gemini. The decision says enhancements to the named feature areas must be available to third parties when Google makes them available to its own AI services.
This will be watched closely outside Europe. Windows has its own emerging AI-agent model through Copilot, local models on Copilot+ PCs, Windows App Actions, and app-level integrations, while Apple is facing similar European pressure over interoperability. Brussels is effectively setting a test case: whether a platform owner can be required to open the system hooks needed for AI agents without turning the phone into a permission nightmare or degrading the security model.
Google’s response over the next year will determine whether the DMA produces genuinely competitive assistants on Android or only a carefully controlled set of APIs that rivals can technically access but struggle to use. By January 2027, the search-data obligation becomes the first measurable milestone. By July 2027, EU Android users should learn whether “choose your assistant” means a cosmetic default—or a real alternative to Gemini.
For Android users in the EU, the eventual change could be as visible as choosing a different assistant and waking it by voice while the screen is off. For developers, it is more consequential: the Commission says competing services must be able to invoke themselves from system entry points, read user-approved context, take actions inside apps, run tasks in the background, and access appropriate on-device AI resources on terms comparable to Google’s own services.
Google has sharply criticized the decisions. Kent Walker, Google’s president of global affairs, said the requirements could weaken privacy and security protections that currently govern powerful system access and the sharing of search data. The Commission, however, says its decision includes consent, integrity, privacy, and cybersecurity safeguards—and that Google can assess serious risks before releasing data to a particular recipient.
The critical detail is that this is not a fine. It is a specification decision: Brussels has now spelled out what DMA compliance is expected to look like. Failing to build and ship the required interoperability mechanisms would open the door to a separate enforcement fight.
Android’s assistant slot is becoming a regulated interface
The Commission’s Android decision covers 11 operating-system capabilities which it describes as the building blocks for capable AI services. They include invocation methods, contextual data access, actions across apps and the operating system, on-device models, and background execution.That reaches far past the familiar ability to set a third-party app as a default assistant. Google will have to ensure that a rival service can be activated from the long-press home control or an equivalent navigation gesture, and through an always-listening hotword mechanism. The Commission explicitly says users should be able to call their preferred assistant even when the display is off, the device is idle, or battery saver is enabled.
The order also targets the capabilities that turn a chatbot into an agent. Under the framework described by the Commission, a user-approved assistant could interact with apps to draft and send an email, add a calendar event, create a note, order groceries, or book transportation. It could use structured integrations where apps expose tasks through Android’s App Functions, or complete multi-step workflows through “Computer Control,” an automation mechanism that mimics user interaction in a separate virtual window.
That is the technical territory where Gemini has enjoyed a major distribution advantage. An AI app can be excellent at conversation yet remain a relatively shallow mobile utility if it cannot see relevant on-device context, invoke itself naturally, or execute a task without forcing the user through each app manually. The Commission’s central argument is that Google cannot keep those system-level pathways for Gemini while presenting Android as open to rival AI providers.
The decision also covers access to microphone, camera, screen, speaker, sensor, and location signals under the same consent and awareness conditions applied to Google. It calls for comparable access to on-device AI models—including Gemini Nano models designated as part of the Android operating system—and comparable hardware and background-execution conditions for third-party local models.
That last point matters beyond the headline fight over Gemini. Local models can offer lower latency, offline functions, and less cloud exposure for tasks such as transcription, proofreading, and summarization. If implementation follows the Commission’s direction, an assistant from OpenAI, Anthropic, Microsoft, Perplexity, Mistral, or a smaller European provider would have a clearer route to building richer Android experiences without having to reproduce every underlying system capability itself.
Consent remains the line between interoperability and surveillance
The Commission’s ruling is not an instruction to hand every AI app blanket access to a user’s phone. Its published guidance repeatedly makes consent the condition for sensitive features. Users must explicitly authorize access for an installed assistant, and applications retain a role in determining which of their data and actions are made available through these channels.That is an important distinction for enterprise Android deployments. A corporate mobility team may welcome more capable assistants for note-taking, translation, scheduling, and accessibility, but it will not want an unvetted consumer AI app able to process screen contents or interact with business apps. The operational question will be whether Android, OEM tools, enterprise mobility management platforms, and the rival assistants themselves can present permissions in a way that is intelligible, revocable, auditable, and compatible with managed-device policy.
Google’s objection is therefore not trivial. The company argues that its existing controls include vetting third-party assistants and that forcing access to sensitive system interfaces can expand the attack surface. A poorly designed integration could create fresh routes for data leakage, overbroad automation, malicious prompt injection, or abuse of accessibility-like controls.
But the Commission rejects the idea that security requires Google-exclusive access. Its position is that parity can be delivered with safeguards and that the DMA requires effective interoperability, not a permission model so restrictive that rival products cannot compete. The ruling gives Google room to assess whether a particular data recipient raises serious cybersecurity or data-protection risks, while also making clear that broad safety claims cannot be used as a permanent barrier.
For security teams, the practical result is likely to be more scrutiny of AI-assistant permissions, not less. The arrival of agents with the ability to observe context and act across apps will make Android’s permission prompts, managed configurations, app allowlists, and logging increasingly central controls.
Google Search data is the second front
The Android measures will draw consumer attention, but the second decision may prove just as significant for the AI market. Google must provide eligible online search engines—and AI chatbots that offer search functionality—access to anonymized search data under fair, reasonable, and non-discriminatory terms.The Commission says Google’s previous proposal was ineffective because it eliminated between 90% and 100% of unique queries from the data set and excluded AI chatbots with search features from the pool of recipients. In Brussels’ view, that left prospective competitors unable to obtain a useful signal for improving ranking, retrieval, query interpretation, and other core search functions.
Google Search’s advantage is not merely its interface or index. It is the enormous feedback loop built from queries, clicks, refinements, and the resulting quality improvements. A new search product or AI answer engine can build a strong model and still struggle to match the practical relevance of a system trained by years of web-scale behavior.
The Commission is trying to narrow that gap without releasing identifiable search histories. Its specifications require a multilayer anonymization approach developed with privacy expertise and aligned with draft guidance from the European Commission and European Data Protection Board. The data may only be used to develop and optimize search services, rather than for unrelated advertising, profiling, or model-training purposes.
Recipients will not receive the data for free. The Commission’s decision allows Google to recover the incremental cost of providing access under a defined pricing framework. That choice is designed to avoid turning the decision into a forced transfer of a valuable asset while preventing price or process from becoming another form of exclusion.
A July 2027 test for Android—and a January 2027 test for search
The key deadlines now matter more than the rhetoric. According to reporting by the Associated Press and other outlets, Google must start sharing qualifying Search data in January 2027. The Android interoperability changes are expected to benefit EU users from July 2027, with reporting indicating implementation must arrive through the next major Android release and no later than August 1, 2027.Those dates leave Google, Android device makers, app developers, AI providers, and regulators with a difficult implementation period. The Commission has not dictated every API design, permission screen, certification workflow, or background-processing limit. Google retains engineering discretion, but not discretion to reserve the practical benefits of a newly covered capability for Gemini. The decision says enhancements to the named feature areas must be available to third parties when Google makes them available to its own AI services.
This will be watched closely outside Europe. Windows has its own emerging AI-agent model through Copilot, local models on Copilot+ PCs, Windows App Actions, and app-level integrations, while Apple is facing similar European pressure over interoperability. Brussels is effectively setting a test case: whether a platform owner can be required to open the system hooks needed for AI agents without turning the phone into a permission nightmare or degrading the security model.
Google’s response over the next year will determine whether the DMA produces genuinely competitive assistants on Android or only a carefully controlled set of APIs that rivals can technically access but struggle to use. By January 2027, the search-data obligation becomes the first measurable milestone. By July 2027, EU Android users should learn whether “choose your assistant” means a cosmetic default—or a real alternative to Gemini.
References
- Primary source: Computerworld
Published: 2026-07-17T12:30:42+00:00
Google must open Android to rival AI agents, EU orders – Computerworld
The move has implications for CISOs as well, says an industry expert.
www.computerworld.com
- Related coverage: digital-markets-act.ec.europa.eu
Commission provides guidance to Google for AI interoperability on Android and sharing of Google Search data under the Digital Markets Act
Commission provides guidance to Google for AI interoperability on Android and sharing of Google Search data under the DMAdigital-markets-act.ec.europa.eu - Related coverage: thehackernews.com
E.U. Orders Google to Open Android Mic, Camera and Screen to Rival AI Assistants
EU orders Google to open 11 Android 18 features to rival AI assistants by August 2027, while sharing anonymized Search data under DMA rules.thehackernews.com
- Related coverage: macrumors.com
EU Orders Google to Give Rival AI Apps the Same Android Access as Gemini
The European Commission today formally ordered Google to give third-party AI services the same access to Android device features that Gemini has. Europe's Digital Markets Act requires software makers like Apple and Google to grant equal interoperability to third-party apps and services, and...www.macrumors.com - Related coverage: washingtonpost.com
- Related coverage: 2eu.brussels
Commission orders Google to open more Android functions to rival AI assistants and share Search data
From July 2027, Android users in the European Union should be able to activate a rival AI assistant by voice and allow it to perform tasks in apps, such as booking a taxi. From January 2027, Google must give eligible search engines and chatbots paid access to anonymised data used to improve its...2eu.brussels