EU orders Google to open Android's AI layer to Gemini rivals

Two binding DMA specifications hand ChatGPT, Claude and other assistants the same system-level Android hooks Google reserves for Gemini — and force anonymized Search data sharing from January 2027.

The European Commission on Thursday issued two binding specification decisions under the Digital Markets Act that pry open the AI layer of Android and, separately, compel Google to share anonymized Search data with rivals. Roughly 427 million EU handsets, per Omdia figures cited by CNN, sit inside the regime.

The first decision names 11 Android system-level features that must be exposed to third-party AI agents by the next Android release in July 2027. The list covers the hooks Google has quietly reserved for Gemini: wake-word voice invocation, in-app actions like booking a taxi or drafting a chat reply, and queries against a user’s recently visited places. ChatGPT, Claude and anything else with an assistant surface get the same reach.

The second decision, effective January 2027, forces Google to share anonymized Google Search data with rival search engines and AI chatbots, using a multi-layered anonymization method developed with the European Data Protection Board and priced under what the Commission calls a fair formula. Google may pre-screen recipients for cybersecurity and data-protection risk.

Google is unhappy. Kent Walker, Google’s president of global affairs, warned that “Today’s decisions risk undermining vital privacy and security guardrails for millions of Europeans,” and said the Search-data mandate would “weaken citizen privacy, risk business trade secrets and endanger national security.” Android president Sameer Samat posted on X that the Commission “is on the wrong track.”

Apple, which pulled its new Siri AI from EU iPhones and iPads in June citing the DMA, told CNN the rules would “give any virtual assistant direct access to users’ private data.”

The enterprise read is the one worth sitting with. Roman Stanek, CEO of Good Data AI, noted that mobile security posture assumes one privileged assistant per device. “Once multiple agents get equal system-level reach… that assumption breaks,” he said, urging CISOs to govern agents “like they govern app stores and MDM policies today.”

Brussels has decided the assistant layer is infrastructure. Once a regulator classifies a surface that way, the incumbent’s product-design privileges stop being privileges.