In Europe, security has become the price of entry to the AI market

COMMENTARY: On July 16, the European Commission issued two binding decisions setting out how Google must comply with the Digital Markets Act (DMA).

The first opens 11 Android features to rival AI assistants. A user can now activate an assistant such as ChatGPT or Claude by voice and run it at the operating system level, as Gemini runs today, rather than inside an app. The second requires Google to share anonymized ranking, query, click and view data from Google Search with competing search engines, including AI chatbot providers, on fair, reasonable and non-discriminatory terms.

[SC Media Perspectives columns are written by a trusted community of SC Media cybersecurity subject matter experts. Read more Perspectives here.]

Data sharing begins in January 2027 and the Android changes reach users from July 2027.

Two clarifications will save readers some confusion. There’s no fine here, and the EU AI Act is nowhere in sight. These decisions conclude a six-month dialogue in which the commission tells a designated gatekeeper like Alphabet precisely how to meet obligations it already carried under those competition laws. That makes this the clearest example yet of AI competition policy, a category most governance discussions have not dealt with yet.

Regulation of AI so far has concentrated on how models behave: bias, hallucination, copyright, misuse. Brussels has now turned to a different question: Who controls access to AI users? As assistants become the front door to search, apps and services, distribution will decide the AI market as much as model quality will.

The commission's case rests on an asymmetry. Install a rival assistant on an Android phone and we get an app. Accept Gemini and we get an operating system feature that answers to our voice, reads our screen and works across other applications.

Alphabet owns the platform on which billions of people will first meet AI, and the DMA forbids a gatekeeper from reserving that position for its own services. The search data decision follows the same logic. Google improves search using behavioral data at a scale no rival can match, so the commission wants a defined, anonymized slice of it available to competitors, giving alternatives to Google Search and Gemini a realistic chance of existing.

Google argues the decisions weaken privacy and security guardrails for millions of Europeans, expose private searches to unfamiliar companies and hand external software sensitive device permissions. These are not small matters. Granting privileged access to more parties expands the attack surface of the world's most widely-used mobile platform, and recent history shows what privileged third-party software can do without an attacker anywhere in sight.

The CrowdStrike outage of July 2024 grounded flights and stopped hospital systems because a single faulty update ran with deep access to millions of Windows machines. Every AI assistant granted system-level access to Android becomes a comparable dependency.

The commission has created a set of specific controls rather than a general assurance. Search data will pass through a multi-layered anonymization process designed with privacy specialists, including stripping direct identifiers and suppressing rare queries that could single someone out. Recipients face contractual limits on how they use and store the data, plus audit requirements, and Google may charge for access at cost with a reasonable return.

Most significantly, Google is permitted to assess, before sharing anything, whether a specific recipient poses a serious cyber security or data protection risk, and to withhold data where it does. The Android measures likewise carry safeguards for user privacy, device integrity and security within the specification.

Security becomes a condition of market access

It’s important to focus-in on the last provision. The commission can deny a company's access to Europe's dominant mobile ecosystem – and to the search data that feeds AI products – on security grounds. Security has stopped being a control applied after the commercial decision and become part of the qualification to compete at all. The EU has been moving this way for years.

The Cyber Resilience Act, NIS2 and DORA all treat cybersecurity as a condition of participating in the digital economy rather than an obligation bolted on afterwards. Demonstrating security maturity has turned into the price of entry to major digital ecosystems, for AI providers first, and before long for the organizations that connect to them.

This decision also corrects a common misreading of European AI regulation, that it begins and ends with the AI Act. AI in Europe has been governed by a portfolio: the AI Act for trustworthy AI, the DMA for competition, GDPR for personal data, the Digital Services Act for platform accountability, the Cyber Resilience Act for product security, and NIS2 for operational cyber risk. The new competition law dictated AI architecture, specifying which system features must be open, which data must flow, and on what terms.

Security leaders can’t wait for this to arrive as a consumer story. Within 18 months, AI assistants from several companies will sit inside the operating system of most of Europe's phones, including the ones our employees carry, with the ability to hear voice commands, read what’s on screen, and act inside other apps.

We need to handle an assistant with that reach the way we would handle any supplier with deep access to our systems:

Know which assistants are running on devices that touch corporate data. Decide which ones we are prepared to allow, based on the vendor's security record and where the data goes, rather than on whichever one the user happened to pick. Set what it can see and do, watch how that changes with each update, and know how we would remove it if the vendor gets compromised.

None of this is new. It’s vendor risk management applied to a new category of software, and the organizations that extend it to AI assistants now will be in far better shape than those that discover in 2027 that a dozen assistants are already embedded across their estate.

Europe has decided that it will govern AI through market structure as well as model rules. Governing AI inside our organization increasingly means the same thing: manage the ecosystems it arrives through, not just the models we chose to deploy.

Steve Durbin, chief executive, Information Security Forum

SC Media Perspectives columns are written by a trusted community of SC Media cybersecurity subject matter experts. Each contribution has a goal of bringing a unique voice to important cybersecurity topics. Content strives to be of the highest quality, objective and non-commercial.