Google must give rival AI assistants the same access to Android’s voice controls, on-screen context and app actions that its own Gemini already enjoys. The European Commission ordered the change on Thursday, one of the broadest interoperability mandates Brussels has issued against a single company.
Android runs on about 60% of smartphones across the European Union, and regulators built the entire case around that number. The fix carries a catch few of Thursday’s headlines flagged: the openness mandate collides with a separate EU cybersecurity law, and Google still gets to decide which rivals qualify for access.
The Eleven Features Now Up for Grabs
The Commission’s order covers eleven specific features grouped into four categories, the same structure Gemini already uses to operate as part of Android rather than as a downloaded app competing for a home-screen tap.
| Access Category | What Changes | Example |
|---|---|---|
| Invocation | Rival assistants can be woken by a custom voice command or dedicated button, not just an app icon | Saying a competitor’s wake word instead of “Hey Google” |
| Context | Assistants can read on-screen content and recent app activity to answer questions about it | Asking about a place a user looked up earlier that day |
| Actions | Assistants can complete tasks inside other apps, including tasks that keep running in the background | Booking a taxi or drafting a reply in a messaging app |
| Resources | Assistants get access to on-device AI hardware and Google’s local models to run tasks efficiently | Using the phone’s own chip instead of routing everything to the cloud |
The Commission says the aim is to grant AI rivals equal access to Android features that Gemini has held alone. A second, related order requires Google to share anonymized search data, including ranking, query and click information, with competing search engines and AI chatbots starting in January 2027.
That second piece matters because it hands over the raw material search tools need to get better. Google has been racing to expand its own reach in the same direction anyway, folding search into a task layer that acts on the open web rather than just answering questions. The order effectively tells rivals they can chase the same target with Google’s own data.
Six Months, One Binding Order
The case has a paper trail. Regulators opened formal proceedings on January 27, 2026, under case reference DMA.100220, then spent months gathering input before adopting binding rules.
- January 27, 2026: The Commission opens formal specification proceedings covering both Android AI access and search data sharing.
- April 27, 2026: Regulators publish preliminary findings and draft measures, opening a consultation for AI providers and device makers.
- July 16, 2026: The Commission adopts both decisions as legally binding, eleven days ahead of its own six-month statutory deadline.
- January 2027: Google must begin sharing anonymized search data with qualifying rivals.
- July 2027: Google must give rival AI assistants voice activation and cross-app access equal to Gemini’s, with the bulk of the technical work due by August 1, 2027.
The Commission’s own public consultation on Android access rules shows regulators asked AI providers and device makers directly what effective interoperability should look like before writing the final text.
“Thanks to these measures we hope to see emerging alternatives to Google Search and Google’s AI services, such as Gemini, and that users in the EU can enjoy greater choice of services,” said Henna Virkkunen, the Commission’s executive vice president for tech sovereignty, security and democracy.
Two EU Laws Are Pulling in Opposite Directions
Specialists split hard on whether Thursday’s order actually works as designed. Two research groups published critiques in the days before the ruling landed, arguing the interoperability mandate collides with a different EU law entirely.
The R Street Institute, a Washington policy group, argued in a July 15 critique that the DMA’s interoperability mandate runs into direct conflict with the EU’s own Cyber Resilience Act, which requires companies to shrink their attack surface and bear lifecycle liability for security, not expand access to outside developers. The Commission has not said which law wins when they clash.
The European Centre for International Political Economy raised a related point in a paper describing the new invocation layer as modeled on Circle to Search, Google’s own feature for pulling up search results from anywhere on screen. That system works as a persistent overlay that can be invoked from designated access points, and handing the same power to outside developers changes what counts as a secure device.
ICLE, a law and economics think tank that filed formal comments during the consultation, went further. It warned Google might weaken AI features instead of opening them on equal, less-vetted terms, calling the resulting package an attack surface qualitatively different from anything the DMA’s interoperability article had previously been used to open.
Today’s decisions risk undermining vital privacy and security guardrails for millions of Europeans.
Kent Walker, Google’s president of global affairs, made that statement after the ruling. He said Google had repeatedly offered solutions to safeguard users while satisfying the DMA’s goals, and that the rulings discount extensive evidence of user harm.
Where the Experts Land
- R Street Institute – warns the mandate directly conflicts with the EU’s Cyber Resilience Act, which pushes companies to shrink their attack surface rather than open it.
- ICLE – told the Commission the rules could push Google to weaken or withdraw AI features in Europe rather than share them under equal, less-vetted terms.
- Fiona Scott Morton, Yale economist and Bruegel fellow – argues the four access categories are the right ones to stop Gemini from locking up the market before rivals get a real foothold.
Google Keeps the Keys to Its Own Gate
Even after losing exclusive access, Google does not lose control of the front door. The rules let the company screen every applicant before granting entry.
- Google can assess any applicant for cybersecurity and data-protection risk before granting access, across five sensitive areas identified in the rules.
- Search records with rare or unusually identifying details get suppressed before sharing.
- Users are grouped into bundles of at least 1,000 so no single search history can be isolated, a threshold The Register first reported.
- Google sets the price formula for the shared data, subject to Commission oversight.
On Google’s own Pixel phones, Gemini is switched on from first setup, the exact level of built-in access competitors have spent years asking Brussels to match. The irony is that Google, the company being forced open, still runs the eligibility check for everyone trying to walk through.
Which Assistants Benefit First?
ChatGPT and Claude are best placed to move fast, according to analysts who reviewed the certification requirements: both already have large European user bases and the engineering staff to clear Google’s approval process quickly. European challengers such as Mistral’s Le Chat could follow, gaining a distribution channel that has effectively belonged to Google alone.
- ChatGPT (OpenAI) – has the scale and EU user base to justify the certification costs, according to policy analysts who reviewed the rules.
- Claude (Anthropic) – named alongside ChatGPT as one of the assistants best positioned for early access.
- Mistral’s Le Chat – a Paris-based challenger that could gain the voice-activation and cross-app rights Google currently reserves for Gemini.
- Perplexity – a search-focused assistant positioned to benefit from both the Android access rules and the separate search-data order.
Fiona Scott Morton’s assessment carries a condition worth keeping in view: her paper argues the measures would help prevent the AI market from locking up around Gemini only if enforced as adopted, not as eventually negotiated down.
Apple Already Tried the Alternative
Google is not the first gatekeeper to face this exact demand. Apple went through its own Article 6(7) case and chose a different path entirely: it walked away.
Apple proposed a Trusted System Agent designed to let outside assistants reach the same device capabilities as Siri AI while preserving what it called privacy and security. EU officials declined every version of that proposal. Apple has said Siri AI will not launch in the European Union alongside iOS 27, meaning EU iPhone owners will not get upgrades available everywhere else, at least for now.
Google took the opposite approach on timing. It launched Gemini’s deep Android integration first and is negotiating compliance afterward, a sequencing that has let EU Android users keep full Gemini access while the two-year case worked through Brussels. The order also lands on top of an existing €4.1 billion fine the Commission issued against Google for favoring its own services on Android, part of a broader run of enforcement that has also pushed Meta to strip out infinite scrolling and pressed Google to open its Play Store.
The Hardware Escape Hits a Lawsuit
Some AI companies are trying to dodge the entire platform-access question by building their own hardware instead of living inside someone else’s operating system. OpenAI hired Jony Ive, Apple’s former design chief, to lead that effort last year, buying his startup io Products for roughly $6.5 billion.
That plan just got more complicated. Apple sued OpenAI on July 10 in federal court in Northern California, alleging its trade secrets were stolen to help build OpenAI’s own consumer devices. The complaint names OpenAI, io Products, hardware chief Tang Tan and former Apple engineer Chang Liu as defendants, and it claims the misconduct reached OpenAI’s senior leadership. OpenAI has denied the allegations, saying it has no interest in other companies’ trade secrets.
The filing also points to more than 400 former Apple employees who now work at OpenAI, a detail Apple used to argue the alleged theft goes well beyond two named engineers. Ive himself is not named as a defendant. The lawsuit does not touch the separate, older partnership that still puts ChatGPT inside Apple’s own products today.
If Google misses either deadline set this week, the Commission can fine it up to 10% of its global annual turnover, rising to 20% for repeat infringements, stacked on top of the Android fine it is already contesting.
Frequently Asked Questions
Does the Android order also cover Apple’s Siri?
No. Apple is working through a separate Article 6(7) case. Google was designated a DMA gatekeeper in September 2023 alongside eight core services, including Search, Android, Chrome, Play, Maps and YouTube, and has been fully bound by DMA obligations since March 2024; Apple’s own case followed a different track that ended in Siri AI’s withdrawal from the EU rather than compliance.
Can Google appeal the decision?
Google has not said whether it will challenge the ruling. EU specification decisions of this kind are enforceable immediately once adopted, and a challenge would go to the General Court in Luxembourg, a process that can run for years without automatically pausing the underlying obligation.
Will Android users have to do anything to use a rival assistant?
Likely yes. Regulators have separately scrutinized Android’s existing choice-screen settings, and early reporting suggests users will need to actively select a preferred assistant rather than have one appear automatically, similar to how default browser and search choices are already presented on Android.
Does the ruling change anything for Android users outside the EU?
Not directly. The order only binds Google’s conduct inside the European Union. Android users in the United States, United Kingdom or elsewhere will not get mandated changes from this decision, though Google could choose to extend similar access globally on its own.
How does this compare to earlier EU interoperability orders?
It goes much further. Earlier Apple iOS specification decisions covered narrower connectivity functions such as NFC, Bluetooth pairing and notification forwarding. Analysts at ICLE described the Android AI package as opening a materially larger attack surface than any previous use of the DMA’s interoperability rules.
