The European Commission on Thursday issued two landmark binding specification decisions under the Digital Markets Act (DMA), compelling Google to fundamentally alter how its Android operating system and Search services operate within the European Union. These rulings, adopted on July 16, six months after proceedings commenced on January 27, aim to level the playing field for rival artificial intelligence (AI) assistants and search engines by dismantling perceived gatekeeper advantages. Google now faces stringent deadlines, with the most significant changes slated for the next major Android release, Android 18, and no later than August 1, 2027.
The Digital Markets Act (DMA) Framework: A Push for Fairer Competition
At the heart of these decisions lies the Digital Markets Act, a seminal piece of EU legislation designed to ensure fair and contestable digital markets. Enacted to curb the power of large online platforms deemed "gatekeepers," the DMA imposes specific obligations and prohibitions on companies like Google, which exert significant influence over the digital economy. Gatekeepers are defined by criteria such such as market capitalization, number of active users, and control over core platform services. With Android commanding approximately 60% of European mobile users and Google Search dominating the search market, Google’s designation as a gatekeeper was unequivocal, triggering the current regulatory scrutiny.
The proceedings initiated by the Commission are not punitive fines but rather "specification decisions." This means the Commission is dictating what a gatekeeper must build or change to comply with the DMA’s principles of interoperability and data sharing. Failure to comply with these specifications, however, could trigger separate non-compliance cases, which carry the potential for substantial fines, underscoring the gravity of these mandates.
Unlocking Android: A New Era for AI Assistants
The first of the two binding decisions directly addresses Google’s Android ecosystem, mandating unprecedented access for rival AI assistants. Historically, Google’s proprietary AI assistant, Gemini (formerly Google Assistant), has enjoyed deeply integrated access to Android’s core functionalities, offering a significant competitive edge. The Commission’s ruling aims to dismantle this exclusivity, requiring Google to grant third-party AI assistants "the same reach into Android that Gemini already has."
This comprehensive access includes critical device components and data streams:
- Camera and Microphone: Enabling rival assistants to process visual and audio input directly.
- On-Screen Content: Allowing assistants to understand and interact with whatever is displayed on the user’s screen.
- Wake Word Functionality: The ability for third-party assistants to activate even when the display is off, mimicking Gemini’s always-on listening capabilities.
- Background App Control: The capacity to drive other applications in the background by simulating user interactions like taps and typing, facilitating complex multi-app workflows.
The Android decision meticulously details 11 operating system features, categorizing them into "restricted" and "unrestricted" based on whether they require certification for third-party access.
Restricted Features (Requiring Certification):
Google is permitted to demand certification before an app can access five highly sensitive features. These "restricted features" include deeply integrated interactions with Google’s own application suite, providing a level of utility previously exclusive to Gemini. A certified third-party assistant will be able to:
- Retrieve and draft emails within Gmail.
- Create and manage events in Google Calendar.
- Access content from Google Drive and Docs.
- Trigger navigation in Google Maps.
- Control YouTube playback and query watch history.
- Read and write SMS, MMS, and RCS messages via Google Messages.
- Initiate phone calls.
This level of integration signifies a profound shift, transforming Android into a more open platform where the user’s chosen AI assistant can orchestrate a wide array of tasks across different applications, regardless of their developer.
Unrestricted Features (No Certification Requirement):
Six other features are to be opened to all third parties, including user-installed apps, without any certification requirement. Google is explicitly barred from restricting the type or use case of the app calling these features. These include:
- Ambient Data: Continuous background access to microphone input, system audio, camera, screen contents, location, and various sensors (e.g., accelerometer). This access must be granted under the same consent prompts as Google’s own AI services, a significant point given that Google’s first-party services currently benefit from "reduced consent processes and privacy indicators," while third parties often face "runtime consent per use."
- Always-on Hotword Detection: The ability to detect a wake word using the low-power Digital Signal Processor (DSP), allowing the assistant to remain active even with a locked screen or in battery saver mode, and to continue recording until the user ends the request. This feature is limited to phones equipped with the necessary chip.
- Long-Press Invocation: Activating an assistant via a long-press gesture.
- System-Level On-Device Models: Access to the underlying on-device AI models.
- Third-Party Model Implementation: The ability to integrate and run third-party AI models.
- Background Execution: Allowing assistants to perform tasks in the background.
The Commission acknowledges the complexity of concurrent hotword detection, where multiple assistants might simultaneously listen for their respective wake words. To allow for necessary technical development, this specific capability will be rolled out later, with a deadline of Android 19 and August 1, 2028.
Crucially, user consent remains paramount for all these features. However, for the six unrestricted features, Google cannot dictate who is allowed to ask for access. While Google can still require process isolation and encryption, any attempt to gate the raw sensor feed behind a new requirement would necessitate a reasoned request to the Commission demonstrating good cause.
The Qualified AI Assistant Programme (QAAP): Google’s New Mandate
For the five "restricted" features, Google is mandated to establish a "Qualified AI Assistant Programme" (QAAP). This programme will enable third-party "Trusted Certification Authorities" (TCAs) to certify assistants, free of charge. Google must accept these certifications without imposing additional conditions and cannot revoke them.
However, Google retains a significant role in defining the QAAP’s terms and approving TCAs. These terms must be "reasonable and non-discriminatory" and cleared with the Commission two months before any changes are implemented. While Google cannot revoke an assistant’s certification once granted by a TCA, it can revoke the certification of a TCA itself, providing a crucial lever of control over the programme’s integrity.
The certification bar itself is capped, focusing on core safety and security. Google may only test whether an assistant:
- Reconfirms user intent before sensitive or irreversible actions.
- Minimizes inadvertent data disclosure.
- Clears baseline mobile app security standards.
- Is hardened against "agentic risks" that could negate user intent, with examples including input, supply chain, integration, model integrity, and infrastructure risks.
Any additional certification requirements beyond these must first be approved by the Commission, and the same stringent conditions must apply equally to Google’s own Gemini assistant.
Users are also provided with a direct "door around the gate." Paragraph 135 of the decision explicitly allows users to consent their way out of the certification requirement on a per-service, per-device basis, ensuring this switch is not buried in developer settings.
The deadlines for the QAAP are tight: draft terms are due by February 1, 2027, with final terms and open applications by May 1, 2027.
Implications for Android App Developers:
By August 2027, developers need to be prepared for certified or user-approved assistants to interact with their apps in unprecedented ways. An assistant could open an app on a virtual display, read its screen, and click through it while the user is engaged elsewhere. Developers will have the functionality to allow their controlled app to block sensitive views from the controlling assistant, a decision that should be wired into their applications before the Android 18 beta. While the decision permits Google to build features for blocking automation on parts of an app or keeping an app’s context away from proactive suggestion components, it does not require Google to do so, leaving some uncertainty for developers.
Democratizing Search Data: Fueling Competition
The second significant decision targets Google’s Search dominance, mandating the sharing of anonymized Search query, click, and ranking data with rival search engines and AI chatbots that perform search functions. This data sharing will be provided for a cost-based fee, aiming to equip competitors with valuable insights currently monopolized by Google, thereby fostering innovation and greater competition in the search market.
The Anonymization Method:
The Commission has prescribed a rigorous, multi-pass anonymization method to protect user privacy while still providing useful data:
- Direct Identifier Stripping: Removal of direct identifiers and attributes that could allow records to be stitched back together, such as usernames, IP addresses, precise timestamps, and input formats.
- Suppression of Rare Terms: Any record containing queries with rare terms (e.g., full names, passwords, street addresses, bank account numbers) or unusually long queries will be suppressed.
- Metadata Generalization: Metadata will be generalized until every user is grouped with at least 1,000 others sharing location, device type, and query language. A further safeguard ensures 95% of users land in groups of 29,000 or more.
Contractual Safeguards and Recipient Criteria:
Access to this anonymized data comes with strict contractual obligations:
- Ringfenced Processing: Data must be processed in isolation, without linking to other datasets.
- No Onward Disclosure: Prohibition on sharing the data with other parties.
- No Re-identification Attempts: Explicit ban on any efforts to re-identify users.
- Audits: An independent audit is required before initial access and annually thereafter.
Recipients of this data must meet specific criteria:
- User Threshold: A minimum of 50,000 monthly average EU users over the past year.
- No Sanctions: The entity cannot be sanctioned.
- Country Risk Exclusion: The entity cannot be controlled by a country the EU treats as a serious and structural cybersecurity or data protection risk.
The data provided will be at least seven days stale and will cut off after five years per beneficiary. Furthermore, Google retains the right to assess, before sharing, whether a specific recipient poses serious cybersecurity and data protection risks.
The timeline for implementing this data sharing is notably aggressive: Google must provide an eligibility form and a beneficiary webpage by the end of August 2026, a finished dataset by November 2026, and a pricing structure by January 2027.
Google’s Stance: Security, Privacy, and Trade Secrets Concerns
Google’s President of Global Affairs, Kent Walker, has voiced strong objections to both decisions, particularly the Android mandate. He stated that the Android decision "threatens device security by granting external apps sensitive and powerful device permissions" and argued that it strips away existing safeguards where phone makers vet AI assistants. Walker highlighted the potential for severe security and privacy risks if third-party apps gain such deep access without sufficient oversight.
Regarding the Search data sharing, Walker contended that the anonymization process is "not good enough," raising concerns that user privacy could still be compromised. He also argued that users are not being adequately asked for consent regarding their data sharing and that the mandate risks exposing Google’s trade secrets and potentially compromising national security. To bolster his argument, Walker cited a recent report from ENISA, the EU’s cybersecurity agency, which emphasized that "security fundamentals matter more than ever in the age of AI." While the ENISA paper itself focused on the rapid vulnerability-exploitation cycle in frontier AI models and did not specifically mention Android, interoperability, or app permissions, Google used it to underscore the broader heightened risk landscape in AI.
Google’s concerns are not entirely unfounded. The decision’s digital-context list grants third-party assistants access to notifications, SMS, screen contents, and screenshots. This is particularly salient given a past incident where SafeBreach demonstrated how WhatsApp and Slack notifications could be used to hijack Gemini’s own Android Utilities agent via indirect prompt injection, requiring no malicious app on the device. Google mitigated this vulnerability server-side in November 2025, prior to SafeBreach’s public disclosure. The Commission’s final measures acknowledge this risk by requiring candidate assistants to be "hardened against agentic risks," for which Google will now write the test.
Evolution of the Mandate: A Compromise Emerges
A crucial aspect of these decisions is their evolution from the initial drafts. The Commission’s draft measures from April 27, 2026, were significantly more expansive in their demands and less accommodating to Google’s concerns. The initial draft for Android, for instance, contained no "restricted features," no "Qualified AI Assistant Programme," and no certification authorities. Draft paragraph 134 explicitly forbade Google from restricting beneficiaries and allowed verification processes only if run by neutral, independent third parties and applied solely to the Play Store. The final decision, in contrast, allows Google to certify applicants itself and includes specific integrity clauses that were not present in the draft.
Similarly, the initial Search draft was looser, lacking a user threshold, capital requirements, or country-risk exclusions, and proposed a metadata group floor of 50 users instead of the final 1,000. Deadlines also shifted, with most Android features originally due by January 1, 2027, now extended to August 1, 2027, and concurrent hotwords pushed from January 2027 to August 2028.
These changes indicate that the final measures represent a negotiated compromise, taking "due account of what Google and third parties filed during the consultation." Google successfully lobbied for the inclusion of a certification regime and the distinction between restricted and unrestricted features, as well as more stringent conditions for Search data recipients and delayed implementation timelines.
However, Google did not achieve complete discretion. The integrity measures it implements must be "strictly necessary," justified by objective evidence it must retain, verifiable by independent parties, and applied identically to its own services. Google cannot impose "a higher level of integrity on third parties than it applies to itself" and must provide the Commission with four weeks’ notice before applying any such measure, unless the change is non-user-facing, purely technical, identical for all, and harmless to third parties.
Broader Implications for the Digital Ecosystem
These decisions mark a pivotal moment for the European digital landscape and set a precedent for DMA enforcement globally. For consumers, the rulings promise greater choice and potentially more innovative AI assistants and search experiences, moving away from a largely Google-centric default. The ability to seamlessly integrate a preferred AI assistant into the Android experience, with deep access to core functionalities and first-party apps, could foster genuine competition in a rapidly evolving AI market.
For rival developers and AI companies, these mandates unlock unprecedented opportunities. Access to critical Android features and anonymized Search data lowers entry barriers and provides essential resources to compete with Google. This could stimulate a new wave of innovation, leading to more specialized, privacy-focused, or regionally tailored AI assistants and search engines.
For Google, the implications are profound. The company faces a substantial engineering challenge to implement these changes across Android and Search, requiring significant resource allocation to build the necessary APIs, certification programmes, and data-sharing infrastructure. While the decisions mitigate some of Google’s initial concerns regarding security, the company will now be under intense scrutiny to ensure its implementation is fair and compliant. The upcoming task of drafting the QAAP terms, in public, knowing they will also apply to Gemini, will be a critical test of Google’s commitment to the new regulatory landscape.
The EC’s assertive stance underscores its commitment to reining in the power of digital gatekeepers and promoting a more open and fair digital economy. As the deadlines approach, the world will be watching to see how these ambitious regulations translate into tangible changes for users, developers, and the competitive dynamics of the global tech industry. The fight, as the original article noted, now moves to February 1, 2027, when Google’s draft programme terms are due.
