The European Commission published its preliminary specifications under the Digital Markets Act today telling Google what it must do to give rival AI assistants the same Android privileges that Gemini currently has. The technical asymmetry is the core of the case: Gemini is invoked by holding the power button or saying "Hey Google," can read on-screen content, interact with other apps, and access system-level functions; a user who installs ChatGPT or Claude on the same device gets an app that has none of those hooks. The Commission's preliminary view is that under the DMA's interoperability and non-discrimination obligations, Google must let third parties register as the system-default assistant, share voice-activation and always-on-listening capabilities, and expose the same integration surfaces into Gmail, Calendar, and other Google apps that Gemini uses natively. Third parties have until May 13 to file feedback on the 29-page specification document, with the Commission's final determination due by end of July. Non-compliance fines under the DMA can reach 10% of annual global sales, which for Alphabet is on the order of $30B at recent revenue levels.
The technical reality of what compliance would require is more interesting than the framing suggests, because the things Gemini does on Android are not all single-API surfaces. Default-assistant routing exists in Android already and would mostly require Google to honour the user's selection without hidden re-prompts; that is enforceable through the spec. Always-on listening on a wake phrase, however, requires access to a low-level audio pipeline that has historically been Google-controlled because of battery and privacy implications, and exposing it equivalently to third parties means defining a Google-equivalent permission tier that Google itself has not formally declared as a separate API. The Gmail and Calendar integration surfaces are the most difficult: today Gemini reads structured user data through internal Google Workspace APIs that are not part of any documented public surface, so equivalent third-party access would require Google to publish (and maintain) a competing-AI integration API. Google's response calling this "unwarranted intervention" is partly accurate to the engineering effort required and partly the standard pre-litigation positioning. The Commission's leverage is the timeline: a 10-week window from publication to final decision is short enough that Google must engage on the spec rather than litigate before the deadline.
The broader implication is that the DMA is being used as the primary regulatory lever to attack the OS-versus-app capability gap that has emerged with built-in AI assistants, and the precedent matters beyond Google. Apple Intelligence on iOS faces a structurally identical question: Siri is invoked by the same OS-level mechanism, and ChatGPT integration is gated through Apple's own contractual and architectural choices rather than a level playing field for Anthropic, Mistral, or any other rival. If the Commission forces Android open under DMA, the equivalent finding against Apple is the next docket. The same applies to Microsoft's Copilot integration in Windows, although the EU has so far been more focused on Android because of mobile market share. The deeper precedent is that "AI assistant access" is being added to the regulatory category previously occupied by browser-default and search-default fights, with the same DMA enforcement architecture applied. The Anthropic and OpenAI side gain in this is real but conditional on Google actually complying rather than litigating to delay, which the DMA's 10% fine ceiling is designed to deter.
For builders, three concrete things change in the near term. First, if you are building a voice or assistant product targeting Android in Europe, the threshold for treating ChatGPT or Claude integration as a real distribution channel just became lower; the question is no longer "will Google ever open this" but "will the timeline for the EU enforcement match my product launch window." Plan for a 12-18 month horizon rather than indefinite. Second, the engineering work to integrate with the Android default-assistant surface is non-trivial and is largely undocumented today; if the DMA forces Google to publish a real API rather than a marketing-friendly partial implementation, the early movers who build against it will have a meaningful integration depth advantage over later entrants. Third, the EU jurisdictional carve-out matters; whatever Google ships for Europe under DMA pressure will not necessarily ship in the US or APAC unless local regulators force the same outcome, which means cross-region products will need to handle the integration as a regional feature rather than global. The honest builder framing is that this is not a 2026 product story but a 2027-2028 distribution-architecture story, and the May 13 to end-July window is the inflection point that determines whether OS-level AI on mobile becomes a regulated marketplace or stays a Google-controlled platform with thin third-party apps.