Europe Just Ordered Google to Let Rival AI Assistants Say "Hey Google." Hassan Taher Explains Why It's About More Than Models

Hey Google

Table of Contents

Key Takeaways

The European Commission’s July 2026 measures target Google’s control over AI distribution on Android and access to Google Search data, not simply AI model performance.
Rival AI assistants will gain access to important Android capabilities, including voice activation and the ability to perform certain actions inside apps.
Google will also be required to provide qualifying third-party search engines and AI chatbots with anonymised Search data under a defined access and pricing framework.
The measures could strengthen AI competition, but security, privacy, data usefulness, and implementation quality will determine how meaningful the changes become.
The broader lesson for AI companies is that winning the assistant market may depend as much on invocation points and data access as on having the most capable model.

On July 16, the European Commission issued two sets of binding specification measures to Google under the Digital Markets Act. The first requires Google to give competing AI assistants the same access to Android features that its own Gemini enjoys. The second specifies how Google must share Google Search data with third-party search engines – and explicitly extends eligibility to AI chatbots that offer search functionality.

Taken together, these are the most consequential AI competition decisions any regulator has issued to date, and they target a layer of the market that receives far less attention than model capability: distribution.

What the Android Order Actually Changes

The Commission’s reasoning starts from a straightforward observation. On Android phones today, competitors’ AI assistants have restricted access to key operating system functionality, while Google’s own AI services have full access. Roughly 60 percent of EU users are on Android devices. A rival assistant that cannot be summoned by voice, cannot act inside applications, and cannot access the context a user is currently in is not a competitor in any meaningful sense – it is a chat app with worse placement.

The decision changes that. Users will be able to activate their preferred AI assistant via voice command, in the way “Hey Google” works today. Third-party assistants will be able to perform actions in apps on a user’s behalf – the Commission’s examples include booking a taxi, generating suggested replies in chat applications, and answering questions about a recently visited place. Reporting indicates eleven Android features must be opened, with changes reaching users from July 2027. The measures include safeguards intended to protect user privacy, device integrity, and security.

The search data decision is the less discussed and possibly more far-reaching of the two. It addresses why Google’s previous data-sharing offers had proven ineffective in practice, specifying that subject to anonymisation, Google should share the same data it collects to optimise its own search services. The Commission developed a multi-layered anonymisation method with internal and external privacy experts, aligned with draft joint guidelines on the interplay between the DMA and GDPR produced with the European Data Protection Board. It also sets out a pricing formula for the shared data and a transparent access process. Google may assess, before sharing, whether a particular recipient poses serious cybersecurity or data protection risks. Reporting indicates sharing must begin in January 2027. Failure to comply can trigger a separate non-compliance case carrying penalties of up to 10 percent of worldwide annual revenue.

Android smartphone AI assistant user — photo credit: Rawpixel

The Bottleneck Was Never Capability

Hassan Taher, an AI analyst and author who advises organizations on AI strategy and platform dynamics, has argued that the industry has consistently misread where the assistant competition is actually decided. “There has been an assumption that the best model wins the assistant market, and it has never been true,” he has noted. “Assistants are won at the point of invocation. The product a user reaches by reflex – the one attached to the button, the wake word, the lock screen – accumulates the interaction data, the habit, and eventually the default status. Model quality determines whether users stay. Distribution determines whether they ever arrive. Those are different fights, and only one of them is settled by benchmarks.”

Apple’s decision to rebuild Siri around Google’s Gemini rather than ship its own frontier model is the clearest recent demonstration of that logic – Hassan Taher on what WWDC 2026 revealed about the AI race reads the move as a company protecting the surface rather than the model. Apple concluded it could rent capability and keep the invocation point. The Commission has now decided Google should not be able to keep both.

That framing explains why this decision matters more than a typical interoperability ruling. The generative AI market has spent three years competing on capability while the distribution layer remained under the control of a small number of operating system and browser owners. A regulator has now intervened directly at that layer – not by fining past conduct, but by specifying the technical access competitors must receive.

The search data provision compounds the effect. Search query data at Google’s scale is not merely useful for building a search engine; it is a description of what people actually want, refreshed continuously. Extending access to AI chatbots with search functionality acknowledges something the market has already figured out: conversational assistants and search engines are converging into the same product category, and the data advantage that mattered in one now matters in the other.

The Objections Deserve a Fair Hearing

There are serious arguments against this approach, and they are not merely self-interested.

The security concern is real. Every operating system capability opened to third parties expands the attack surface, and an assistant granted permission to take actions inside applications on a user’s behalf is a high-value target. The Commission has built in safeguards and allows Google to assess recipient risk before sharing data, but the honest position is that the risk is being managed rather than eliminated. Anonymisation of search data is likewise a genuinely hard technical problem – the Commission has explicitly reserved the right to amend the anonymisation requirements based on future market developments and independent third-party evaluation, which is a tacit acknowledgment that the initial method may not be sufficient.

There is also a legitimate innovation argument: companies that invest in building integrated platforms may reasonably object to being required to hand competitors the benefits of that integration, and mandated openness can dampen the incentive to build tightly coupled products in the first place.

Taher has cautioned against reading the decision as a straightforward win for competition. “Mandated access changes who the gatekeeper is; it doesn’t necessarily eliminate gatekeeping,” he has observed. “If rival assistants build on shared Google search data and Android hooks, they are competing on Google’s rails, on terms a regulator adjusts periodically. That may still be better for users than the status quo. But a company whose product depends on a compliance regime remaining favorable has a strategic dependency, not independence. Anyone building on these openings should treat the access as a channel, not a foundation.”

What to Watch

Three things will determine whether these measures produce real competition or procedural compliance. The first is implementation quality – whether the eleven opened Android features work reliably and with acceptable latency for third parties, or technically satisfy the order while remaining awkward enough that users stay with the default. The second is whether the anonymised search data retains enough signal to be genuinely useful. Heavily anonymised data can be compliant and worthless simultaneously. The third is whether any of this arrives in time to matter. Assistant defaults are being established now; the Android changes reach users in July 2027 and data sharing begins in January 2027.

For businesses outside Europe, the practical significance is that the EU is again setting a template that tends to travel. The commercial version of the same shift is already underway on the other major mobile platform, where Hassan Taher on Apple turning the iPhone into an AI marketplace describes a platform owner opening the assistant layer by choice rather than by order. Europe is arriving at a similar outcome through regulation; the interesting question is which route produces better terms for the companies building on top.

For anyone building an AI assistant, the strategic message is the one the Commission implicitly endorsed: the fight for the assistant market is a fight over invocation and data access, not over leaderboard position. Europe has decided that fight is too important to leave to the company that currently owns both.

Data sharing compliance — photo credit: Rawpixel

FAQs

What did the European Commission order Google to do?

The European Commission issued binding measures requiring Google to provide competing AI assistants with greater access to Android functionality and to share qualifying Google Search data with third-party search providers, including AI chatbots with search capabilities.

Why does Android access matter for AI assistants?

Access to features such as voice activation and in-app actions can determine whether an AI assistant becomes part of a user’s everyday behavior rather than remaining an application that must be opened manually.

Will Google have to share all of its Search data?

No. The measures establish a framework for sharing qualifying data subject to anonymisation, access conditions, pricing rules, and safeguards concerning cybersecurity and data protection.

Could the EU’s measures create new security risks?

Potentially, because opening additional Android capabilities to third-party assistants can expand the attack surface, although the Commission has included safeguards intended to address privacy, security, and device-integrity concerns.

What does the decision mean for AI companies?

It reinforces the idea that AI competition is increasingly about distribution, default access, invocation points, and data as well as model capability.

Sources: