EU orders Android to open key functions to rival AI assistants

The choice of assistant is moving deeper into the phone

The European Commission has told Google to open important parts of Android to competing AI assistants.

The measures are binding. They cover 11 operating-system functions that the Commission says rival services need if they are to offer something as capable and convenient as Gemini on an Android phone.

This is more than permission to install another chatbot. A person should be able to wake a chosen assistant with a voice command, give it relevant context from the device and let it carry out an agreed task across apps. Today, much of that deeper access is reserved for Google's own services.

The decision was adopted under the Digital Markets Act, the EU law that places extra obligations on large digital platforms designated as gatekeepers. It follows proceedings opened in January and a public consultation on draft measures in the spring.

The change will not appear overnight. Most of the Android work is due in the next major release, Android 18, and by 1 August 2027 at the latest. Support for several assistants to listen for their own wake words at the same time is due with Android 19, by 1 August 2028.

What rival assistants should be able to do

The Commission groups the 11 functions into four practical areas: being invoked, understanding context, acting on apps and the operating system, and getting enough local resources to do the work.

A rival assistant should be able to respond to a long press or a wake word. With permission, it should be able to read context that apps have made available, use live inputs such as the screen or microphone, and offer a relevant suggestion without waiting for a new prompt.

It should also be able to complete structured actions in apps, such as drafting an email or scheduling a meeting. For tasks that do not have a clean integration, the measures cover screen automation in a separate background window. System controls and background execution are included too.

The decision also reaches on-device AI. Third parties are meant to receive equal access to models that form part of the designated Android system, including Gemini Nano models, and to run their own local models under the same hardware and background conditions as Google's services.

Google must provide documentation, testing and technical help. Access to these Android functions is to be free, equally effective and available across the Google Android ecosystem, including devices made by other manufacturers.

That is the Commission's specification. Whether the experience feels genuinely equal will only become clear when developers test the finished interfaces on real phones.

Search chatbots get a second opening

A separate decision deals with a different advantage: the search data Google collects at a scale its competitors cannot match.

Under the DMA, Google already has to share anonymised ranking, query, click and view data with eligible search engines on fair terms. The new measures clarify that AI chatbots with online search functions can qualify too.

The Commission says Google's earlier proposal excluded those chatbots and removed between 90% and 100% of unique queries from the dataset. It says there was no meaningful uptake by potential recipients.

The revised framework tells Google to share broadly the same categories of data it uses to improve its own search service, after anonymisation. Recipients may use it to improve query understanding, retrieval, ranking and indexing. They may not use it to train a general-purpose AI model, build unrelated advertising profiles or simply copy Google's results.

This is not a gift of Google's search algorithm. It is controlled access to a changed and delayed dataset that could help a smaller search service understand what people find useful.

The first implementation milestones arrive quickly: an information page and application form are due by the end of August 2026, test samples and draft licence terms by September, the anonymised dataset by November, and a final price offer by January 2027.

Privacy is part of the engineering, not a footnote

Opening device context and search data creates obvious privacy and security questions.

For Android, the Commission says users must explicitly consent to the features an assistant can access. Google may apply objective security conditions to some of the more sensitive functions, such as screen automation and central access to app data. Independent bodies will help certify eligible services.

For search data, direct identifiers and precise timestamps are removed. Rare or unusually long queries can be suppressed, and metadata is generalised so users sit inside large groups. Contractual limits, secure processing and independent audits sit on top of those technical changes.

Those safeguards are detailed, but anonymisation is not magic. Search queries can contain intimate facts, and device context can reveal even more. The Commission says it will review the system and can reopen the decision if evidence shows the protection is no longer adequate.

That continuing review matters. A dataset can be useful precisely because it retains patterns. The job is to preserve enough utility for competition without leaving a route back to the person behind the pattern.

What is confirmed, and what is still open

The confirmed part is the legal decision. The Commission has adopted final specification measures, set implementation dates and described the functions and data Google must make available.

The Commission's case is also a policy claim: it expects this access to create more competition, better services and more choice. That outcome has not happened yet.

Rival assistants still need strong models, careful product design and enough trust to win a place on someone's phone. Search companies must meet eligibility thresholds, pass audits and decide whether the price and altered data are worth it. Google must turn a long legal specification into interfaces that work reliably without weakening device security.

There is a wider question too. Opening a platform can lower one barrier while leaving others in place, including distribution, defaults, compute costs and brand recognition.

Still, the direction is clear. In Europe, an AI assistant is being treated less like an ordinary app and more like a new way into the operating system. The EU does not want that entrance reserved for the company that owns the door.