Skip to content

Brussels Gave ChatGPT and Claude the Android Powers Google Kept for Gemini

DS
LDS Team
Let's Data Science
10 min
The European Commission ordered Google on July 16 to open 11 Android system features, including always-on voice activation and screen reading, to rival AI assistants by August 2027, and to share anonymized search data with competitors starting January 2027. Google says the ruling endangers the security of the 60% of Europeans who carry an Android phone; ignoring it risks a fine that could top $35 billion.

Say "Hey Google" to an Android phone anywhere in the European Union, and the operating system itself wakes up, listens, and acts. Say the equivalent phrase to ChatGPT or Claude on the same device, and nothing happens unless the app is already open on screen.

That gap is not a bug. It is Android's architecture. Gemini runs as part of the operating system, with standing access to the microphone, the screen, and every other app's data. Everything else is a guest, sealed inside its own sandbox.

On July 16, the European Commission decided that gap is illegal.

In two binding decisions issued under the Digital Markets Act, Brussels ordered Google to give rival AI assistants the same system-level access to Android that Gemini has always had, and to start sharing the search data that keeps Google Search ahead of every competing engine and chatbot. Google published an objection within hours. Compliance is not optional.

Gemini Gets Four Kinds of Access No Downloaded App Can Match

The Commission's technical analysis, laid out in its specification decision, breaks Gemini's advantage into four layers that a normal app cannot reach.

The first is invocation. Gemini registers a wake word at the operating system's audio layer, a background process that runs beneath any app's sandbox. A downloaded assistant can only hear a user while its own window is open.

The second is context. Gemini reads from AppSearch, an on-device index that pulls structured data out of Gmail, Calendar, and Drive, plus a continuous "ambient" feed from the camera, screen, and speakers. Third-party apps can only see their own slice of that index.

The third is execution. Gemini calls other apps' internal functions directly, composing an email in Gmail or updating a calendar entry without the user opening either app. A feature called screen automation lets it operate other apps in a hidden background window.

The fourth is resources. Gemini Nano, Google's on-device model, gets priority access to the phone's Neural Processing Unit and is exempt from the battery-saving rules that suspend ordinary apps.

The Commission's ruling requires Google to open all four layers, covering 11 specific features in total, to any AI assistant that qualifies under a new certification program. Most features must ship in Android 18, with a hard deadline of August 1, 2027. One capability, letting more than one assistant listen for a wake word at the same time, is deferred to Android 19 and a 2028 deadline.

LayerWhat Gemini can do todayWhat rivals get by August 2027
InvocationAlways-on wake-word detection below the app sandboxThe same OS-level wake-word registration
ContextReads the aggregated AppSearch index across Gmail, Calendar, DriveAccess to the same aggregated index, not just their own app data
ExecutionCalls other apps' internal functions; runs screen automation in the backgroundThe same structured API calls and background screen automation
ResourcesPriority access to the Neural Processing Unit; exempt from battery throttlingThe same priority hardware access and battery exemptions

The Second Order Targets an Even Older Advantage

The Commission's other decision goes after something harder to copy than an operating system permission: two decades of search queries.

Google Search holds more than 90% of the EU search market, a dominance that has quietly built what engineers call implicit relevance feedback: every query typed, every result clicked, every result ignored, tied to how high that result ranked at the moment someone chose it. A search engine cannot buy that dataset. It has to accumulate it, one click at a time, over years, at a scale only a market leader reaches.

Under the July 16 decision, Google must share anonymized versions of that data, covering query text, device and language metadata, result URLs, and click and ranking data, with qualifying rivals starting in January 2027. The Commission's anonymization method strips direct identifiers, filters out queries containing names or addresses, and requires every remaining query to fall into a group of at least 1,000 similar users before it can be shared.

The decision defines eligible recipients broadly enough to include any AI chatbot that answers questions by retrieving and summarizing live web results, a description that fits ChatGPT and Claude as much as it fits a traditional search startup. Recipients must have real trading history or meaningful capital, tens of thousands of monthly EU users, and pass an independent annual audit before they get access, and they pay for it under a pricing formula fixed for five years.

Google Says the Fix Creates the Problem It Claims to Solve

Kent Walker, Google and Alphabet's president of global affairs, published Google's response the same day the decisions landed.

"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 (Google, July 16, 2026)

Walker's specific worry is what happens once the sandbox walls come down. "This Android ruling threatens device security by granting external apps sensitive and powerful device permissions without these safeguards," he wrote, pointing to a warning from ENISA, the EU's own cybersecurity agency, that "security fundamentals matter more than ever in the age of AI." On the search side, he argued that anonymized queries can still expose "business trade secrets" and "national security" information once they leave Google's servers for an outside company.

An independent analysis published a day earlier by the R Street Institute found real substance behind that objection: the DMA's interoperability mandate and the EU's own Cyber Resilience Act, which requires companies to shrink their attack surface and ship secure-by-default products, pull in opposite directions the moment a third-party assistant is granted screen-reading and cross-app execution power. The Commission's decision allows Google to apply "proportionate" security checks, but does not say who is liable if a certified rival assistant is later hacked, an open question its own certification program is supposed to answer by 2027.

Not every outside expert reads the ruling as a security downgrade. Fiona Scott Morton, a Yale economist and senior fellow at the Brussels think tank Bruegel, argued in a July 1 working paper that the decision correctly identifies the access points that decide whether a rival AI assistant can compete as a genuine device-level product, and that enforcing it would let AI companies win on merit rather than on who already controls Android's distribution.

The Six Months That Produced a Technical Build Specification

JANUARY 27, 2026
The Commission opens two specification proceedings
Formal review begins into whether Google's existing Android and search-data compliance offers meet its Digital Markets Act obligations.
APRIL 2026
Preliminary findings are published
The Commission signals that Google's prior offer, which had stripped out up to 100% of unique queries and excluded chatbots entirely, falls short.
JULY 16, 2026
Final decisions adopted, 11 days ahead of the statutory deadline
Both binding specification decisions are issued. Google publishes Kent Walker's objection the same day.
END OF AUGUST 2026
Alphabet must open the door to applicants
Google must submit an eligibility application form and publish a public page telling potential beneficiaries how to apply.
JANUARY 2027
Search data sharing begins
Pricing terms are finalized and qualifying search engines and AI chatbots can start receiving anonymized query data.
AUGUST 1, 2027
Android 18 must carry all 11 features to certified rivals
The hard deadline for opening the invocation, context, execution, and resource layers to qualifying third-party AI assistants.
AUGUST 1, 2028
The last feature ships in Android 19
Concurrent always-on wake-word detection, letting more than one assistant listen simultaneously, arrives on the final deadline.

What makes this pair of decisions different from a decade of prior antitrust fights is the mechanism. Earlier EU actions against Google, including the 4.1 billion euro Android fine that became final this July after the company's last appeal failed, punished a past violation. This one, built on the DMA's Article 8(2) specification power, skips straight to an engineering calendar: named data fields, a named anonymization threshold, a fixed pricing formula, and a certification deadline, all enforceable the moment it was signed. Rob Bratby, a UK digital-markets lawyer who tracks the case, called the Commission's pace "notable discipline."

An appeal to the EU General Court is expected. Under a sequencing rule the same court set in July in Apple's parallel gatekeeper cases, filing an appeal does not pause the compliance clock. Google has to build toward the deadline while it argues against it.

Why Practitioners Building on Android Should Care Now

For engineers building assistants, agents, or search products, the practical question is not whether Brussels can order this. It is what "equally effective" access actually looks like once Google writes the certification terms due in February 2027, and whether that program becomes a genuine on-ramp or a slow-walked gate.

The parallel worth watching is Apple's own move to open iOS to third-party AI extensions, which arrived under different pressure but points at the same underlying shift: the assistant a phone maker bundles by default is losing its structural advantage over the one a user actually prefers. The EU's approach, forcing interoperability through a detailed technical mandate rather than trusting a company's voluntary offer, also stands in sharp contrast to the deregulatory instinct driving the Trump administration's own push to preempt state AI laws, which is trying to strip away state-level AI rules rather than add federal ones. Companies increasingly have to build for both postures inside a single product.

It is also the latest example of regulators writing rules with real enforcement teeth, following the pattern seen when Illinois moved to codify AI safeguards into state law rather than waiting on federal action. The difference here is scale: this is not one state, it is the European Union's entire Android and search market, and the fine for ignoring it can reach 10% of Alphabet's global revenue.

The Bottom Line

Strip away the legal architecture and the decision says something simple: an operating system should not be allowed to reserve its best features for the AI assistant it happens to own. Google's counter is just as simple: taking those features away from the sandbox is how phones get compromised.

Both things can be true, and the Commission's calendar does not resolve which one wins. It just starts the clock. By August 2027, an Android phone in the EU either has a genuinely open assistant layer, or it has a new kind of attack surface nobody has finished securing.

As Kent Walker put it: Google will "continue advocating for a balanced approach." Whether Brussels' engineering deadlines leave room for that balance, or simply run past it, is the question every AI assistant maker building for Android now has to answer for their own product.

Sources

Practice with real Ad Tech data

90 SQL & Python problems · 15 industry datasets

250 free problems · No credit card

See all Ad Tech problems