Say "Hey Google" to an Android phone and Gemini answers instantly, even with the screen off. Say anything at all to ChatGPT or Claude on the same device and nothing happens, because Android was never built to listen for them.
That gap was not an accident of engineering. Google's own assistant runs a persistent, always-on listening process at the operating system's audio layer. Every competing AI assistant has been confined to an app: something a user has to open and tap, with no access to whatever else is happening on the phone.
On July 16, 2026, the European Commission ordered Google to close that gap.
The Commission issued two binding specification decisions under the EU's Digital Markets Act. The first forces Google to give rival AI assistants the same system-level access to Android that it reserves for Gemini: the wake word, the ability to read what's on a user's screen, and privileged programming interfaces for acting inside other apps. The second forces Google to share the search data behind its AI-powered search advantage with competing search engines and chatbots, including OpenAI.
Roughly 60% of European mobile users carry an Android device, and Google Search still holds more than 90% of the region's search market, a position the Commission says it has held "for decades." For any company building an AI assistant or an AI-powered search product to compete with Google on equal footing, those two numbers have defined the ceiling. The decisions, adopted six months after the Commission opened its investigations, are an attempt to raise it.
Android Only Ever Listened for One Assistant
Under Article 6(7) of the DMA, gatekeepers like Google must give developers free, effective interoperability with the operating systems they control. The Commission opened a formal proceeding on January 27, 2026, to spell out what that means for Android and AI assistants specifically. Eleven Android features made the final list, grouped into four categories.
| Category | What only Gemini could do | What changes |
|---|---|---|
| Invocation | Wake word and long-press home button access reserved for Gemini | Rivals register their own wake word and home-button trigger |
| Context | Reads on-screen content, microphone, camera, and sensor data | Rivals get the same "ambient data" access, with user consent |
| Actions | Executes tasks inside Gmail, Maps, Messages, and other apps through privileged APIs | Rivals gain equal access to the same on-device app APIs |
| Resources | Priority access to on-device Gemini Nano models and background execution | Rivals can call the same on-device models under equal conditions |
Some specifics explain why this felt like a structural advantage, not a head start. The same long-press of the home button that can summon Gemini also launches Circle to Search, a system-wide gesture reserved for Google; that kind of access point can no longer belong to one company. Gemini also surfaces suggestions nobody asked for, like a flight number pulled from an email while a friend texts about it, through a feature called Magic Cue; rivals get the same proactive pathway. And Gemini automates multi-step tasks inside other apps through Computer Control, opening a virtual window that operates apps in the background; competing assistants gain the equivalent capability, screen automation, once certified.
Certification is the catch. Five of the 11 features, including screen automation and centralized access to on-device app data, are sensitive enough that Google may impose eligibility conditions, verified by independent auditors, before a rival can use them. Google must publish draft certification terms by February 1, 2027, finalize them by May 1, 2027, and decide each application within four weeks.
Google is not being asked to give up Gemini's position inside its own ecosystem. The same models anchoring Gemini on Android also power Siri under Apple's deal with Google, announced in January 2026. What changes is whether Android will still recognize only one assistant as more than an app.
Google's Search Data Becomes a Regulated Product
The second decision targets a different asset: the data behind that search dominance. Scale compounds. More queries improve ranking, better ranking attracts more queries, and no rival has ever accumulated a comparable dataset to catch up.
Under Article 6(11) of the DMA, Google must share anonymized query, click, ranking, and view data with eligible rivals, including AI chatbots that offer their own search features. This is not a new fine or a new law. It is what the Commission calls a specification decision: a binding explanation of how Google must satisfy an obligation it has carried since the DMA took effect in March 2024. The Commission says Google's previous compliance offer stripped out 90% to 100% of unique queries and excluded AI chatbots outright, producing what it calls no meaningful uptake.
The anonymization method is specific. Each user is folded into a group of at least 1,000 people sharing the same location, device type, and query language, a technique called k-anonymity. The Commission expects 95% of users to sit in groups of 29,000 or more, and pricing is capped near Google's own cost of capital: the return it could expect from any similarly safe use of that money, not an open-market rate Google sets itself.
To qualify, a rival search engine needs at least 50,000 monthly average EU users in the past year. A company operating for less than two years can still qualify as a credible new entrant if it has raised more than €50 million in capital.
Google must finalize that pricing and notify eligible rivals by January 2027, the last in a string of milestones that starts with an eligibility application due in late August 2026.
The Compliance Clock Runs Into 2028
Neither decision arrived overnight, and neither finishes quickly. Here is how six months of closed-door proceedings became a compliance schedule running two more years.
Google Calls the Fix a New Kind of Risk
Kent Walker, President of Global Affairs at Google and Alphabet, published the company's response within hours of the announcement. His argument isn't that Google's advantage is imaginary. It's that removing it this way creates a different problem.
"Today's decisions risk undermining vital privacy and security guardrails for millions of Europeans. We have repeatedly offered solutions to safeguard users while satisfying the DMA's goals, but these rulings discount extensive evidence of user harm." — Kent Walker, President of Global Affairs, Google & Alphabet
Walker's objection to the Android decision is that phone manufacturers currently vet any app requesting deep device permissions, and the Commission's order sidesteps that check. "The reality is that AI assistants already safely access Android's capabilities, with phone makers playing a key role in vetting them," he wrote, noting the ruling arrives as the EU's own cybersecurity agency, ENISA, warns that security fundamentals "matter more than ever in the age of AI."
On search data, his objection is narrower: exposure. Opening the dataset to "unfamiliar companies, without adequate anonymisation of the data and without user knowledge or consent," Walker wrote, would risk "business trade secrets" and "national security." Google raised a similar objection in April, when the proposal first went out for comment. Clare Kelly, its senior competition counsel, said the plan would force Google to hand over data behind "private questions about their health, family, and finances... with dangerously ineffective privacy protections."
Google is not the only critic. Daniel Castro, president of the Washington-based Information Technology and Innovation Foundation, said the Commission is "stretching the DMA into areas that were not its intended focus" and "effectively making AI policy on the fly," rather than leaving AI-access decisions to elected lawmakers. Castro went further than Google, warning that continued DMA enforcement against American companies could push US policymakers toward retaliation under Section 301 of the Trade Act.
A more technical objection comes from the R Street Institute, a Washington think tank focused on cybersecurity policy. In a July 15 analysis, the group argued that opening invocation, screen content, and cross-app execution to any qualifying third party conflicts with the EU's own Cyber Resilience Act, which requires gatekeepers to minimize their attack surface and ship secure-by-default products. Neither law, in the group's reading, says which one wins when they point in opposite directions.
The Commission's materials anticipate some of this. Google can still impose vetting conditions on the five most sensitive Android features, and it built the search-data anonymization method with privacy regulators, including the European Data Protection Board. Whether that satisfies Google's actual complaint, that manufacturer vetting is being replaced by something weaker, is an argument regulators and Google will keep having as the 2027 deadlines approach.
Rival AI Assistants Get a Real Path Onto Android
For engineers building an AI assistant or an autonomous agent meant to work on a phone, the gap the Commission just closed is a familiar one. A model can be excellent and still be useless as a phone assistant if the operating system won't let it listen passively, see the screen, or act inside other apps without a person copying and pasting between windows. That has been the practical ceiling for every Android AI app that isn't Gemini.
OpenAI, Perplexity, Microsoft's Copilot, and Samsung's own assistant efforts are the most obvious beneficiaries. Each already ships an AI assistant on Android, and each has been limited, until now, to the same app-sandbox restrictions as any other download. Under the new order, any of them can apply for the wake-word registration, screen-context access, and cross-app APIs Gemini has used to become the default way people talk to their phones. Once certified, a rival assistant on Android can:
- Register a custom wake word that works with the screen off, not just inside its own app
- Read on-screen content to answer questions about whatever a user is looking at, the way Gemini already does
- Send a message, add a calendar event, or edit a shopping list inside other apps through Android's structured APIs, instead of simulating taps on a screen
- Call the same on-device Gemini Nano models Google's own assistant uses, rather than routing every request to the cloud
The search-data decision matters just as much for a narrower set of companies: anyone building a product that answers questions by retrieving and ranking live information from the web, a process often called grounding. Perplexity, which has spent the past year building always-on AI agents, including one it shipped inside a $599 Mac Mini, is a natural candidate once it clears the 50,000-monthly-user threshold. So is any company adding live search to a chatbot.
None of this arrives quickly, and Google isn't required to slow Gemini down while rivals catch up. Apple made a comparable move voluntarily in May, letting iPhone users pick Claude or Grok as alternatives to Siri in iOS 27. Google needed a binding order from Brussels to do the same on Android, and even now, the most consequential changes don't reach a real device until Android 18, expected by August 2027. The version letting multiple rivals each hold their own wake word at once doesn't arrive until a year later.
The Bottom Line
Strip away the legal language and both decisions say the same thing: Google built a phone operating system, then wrote itself a standing exception to its own rules. Gemini could listen, see, and act in ways no competitor was allowed to. Google Search collected decades of query data no rival could replicate. The Commission's response is not a fine. It is a rulebook: 11 Android features, a defined anonymization method, pricing tied to Google's own cost of capital, and deadlines running until August 2028.
Whether that rulebook produces a real competitor to Gemini is a different question from whether it was justified, and lawyers won't answer it. Engineers will, by whether OpenAI, Perplexity, Samsung, or an assistant nobody has heard of yet ships something good enough to make a user say a different wake word.
Google isn't conceding the argument even as it complies. As Kent Walker put it, the company will keep "advocating for a balanced approach." Under the DMA, advocacy and compliance now run on separate tracks, whether Google calls that balanced or not.
Sources
- Commission Provides Guidance to Google on AI Interoperability and Google Search Data Sharing Under the DMA — European Commission, Digital Markets Act, July 16, 2026
- Alphabet Specification Proceedings: Interoperability for AI Services — European Commission, Digital Markets Act, July 16, 2026
- Alphabet Specification Proceedings: Sharing of Google Search Data — European Commission, Digital Markets Act, July 16, 2026
- The DMA Should Not Undercut Security & Privacy for Europeans — Google (Kent Walker), July 16, 2026
- EU Gives Rival AI Assistants System-Level Android Access Google Reserved for Gemini — Tech Times, July 16, 2026
- EU Orders Google to Open Android and Search Data to AI Rivals — MediaNama, July 17, 2026
- Rival AI Assistants Could Soon Gain Full Access to Android Features — Android Authority, July 16, 2026
- EU's Latest Action Against Google Is a Harmful Expansion of DMA, Says ITIF — Information Technology and Innovation Foundation, July 16, 2026
- House of Contradictions: EU's Competition Rules Poised to Undermine Its Own Cybersecurity Law — R Street Institute, July 15, 2026