{"slug": "europe-gets-ready-to-police-frontier-ai", "title": "Europe gets ready to police frontier AI", "summary": "The European Union will gain sweeping powers on August 2 to police large general purpose AI models, allowing the European Commission to demand information, conduct safety evaluations, order corrective measures, pull models off the market, and impose fines of up to 3% of annual total turnover. The rules target providers of the largest models, including OpenAI, Anthropic, Meta, Alphabet, xAI, Alibaba, ByteDance, Z.ai, and Mistral, and enforcement will test whether Brussels is willing to uphold its AI rules against U.S. and Chinese tech giants.", "body_md": "After years of negotiation and regulatory fine-tuning, the European Union will gain sweeping new powers on Sunday to police [large general purpose AI models](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-51) — the versatile, behind-the-scenes AI systems that power the chatbots, image generators and enterprise tools that are how most people interact with AI.\n\nFrom August 2, the Commission will be able to demand information from model providers, conduct safety evaluations, order corrective measures, pull models off the EU market and even impose fines of up to 3% of annual total turnover. The new mandate will place the EU at the global frontier of global AI safety regulation.\n\nBut whether those powers actually matter depend on one question: Will the Commission be willing, and able, to use them?\n\nStrong enforcement would signal that Brussels is ready to uphold its AI rules even if doing so creates friction with the United States or China, home to most leading model providers. Hesitation, by contrast, would reinforce the criticism that the EU is a whole lot better at writing laws than making powerful companies follow them.\n\n“Enforcement depends on the willingness at a political level to enforce the law and take it seriously,” said Risto Uuk, head of European policy and research at the Future of Life Institute. “A lot of stakeholders are watching and hoping that they will be strong from day one.”\n\n**The AI Act enters its enforcement era **\n\nThe August 2 launch has been years in the making.\n\nThroughout much of 2026, the EU moved to simplify its digital rulebook, including the EU AI Act. That culminated in May, when lawmakers [agreed](https://www.theparliamentmagazine.eu/news/article/why-the-eu-rewroteits-landmark-ai-law) to delay rules governing high-risk AI systems, and remove industrial applications from the scope altogether. But they left the timetable for general purpose AI models untouched — a major win for AI safety advocates who argued that regulating the largest, wealthiest and most dangerous parts of the AI ecosystem couldn’t wait.\n\nThe rules target providers of the largest general purpose AI models rather than the businesses that deploy those models downstream — an important distinction, according to Uuk. The threshold for models deemed to pose systemic risk is so high that only a handful of companies, such as the U.S.’s OpenAI, Anthropic, Meta, Alphabet, xAI and [China’s](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6106566) Alibaba, ByteDance and [Z.ai](http://z.ai) are expected to fall within scope. France’s Mistral is the only [European model provider](https://epoch.ai/data-insights/models-over-1e25-flop) that should fall under the law.\n\nObligations under the AI Act for the largest, most powerful models are governed by [Article 55 of the AI Act](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-55), which requires the companies to perform state-of-the-art model evaluations, assess and mitigate systemic risk, report serious security incidents and ensure appropriate cybersecurity protection.\n\nThe AI office has already built much of the infrastructure needed to enforce the rules. Last year, it published the [Code of Practice](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) in collaboration with industry and outside experts, issued [guidelines](https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers) for model providers, established an [AI Act Advisory Forum](https://digital-strategy.ec.europa.eu/en/policies/ai-advisory-forum) to assist with enforcement, partnered with [independent external evaluators](https://digital-strategy.ec.europa.eu/en/events/call-participants-workshop-qualification-requirements-external-evaluators-gpai-models-systemic-risk) like [FAR.AI](http://far.ai) and [Epoch](https://epoch.ai/) AI, and put out a [tender](https://digital-strategy.ec.europa.eu/en/funding/eu-ai-office-launches-eu9-million-tender-technical-support-gpai-safety) for additional technical support on model safety.\n\nTaken together, those preparations leave the EU and model providers in a strong position ahead of August 2, according to Uuk.\n\nHowever, it will also be a test of the Commission's appetite for scrutinizing some of the world’s most valuable companies, most of which are based in the U.S. or China. It took years of pressure from the European Parliament, national leaders like French President [Emmanuel Macron](https://www.euractiv.com/news/macron-slams-eus-slow-enforcement-of-digital-rules-on-us-tech/) and outside groups before the Commission began aggressively enforcing the Digital Services Act and Digital Markets Act, according to Jimmy Farrell, EU AI policy lead at Pour Demain, and Mehmet Onur Cevik, an AI policy analyst at the Centre for Future Generations.\n\n“I am expecting some similarities between what’s happened with DSA and DMA and what will happen with the AI Act,” said Cevik. “These big companies like X, Meta and Alphabet already have enforcement track records with the EU.”\n\nThe AI Office may also [not have enough](https://media.licdn.com/dms/document/media/v2/D4E1FAQFpc7rdRPJzlQ/feedshare-document-url-metadata-scrapper-pdf/B4EZxr0RpaIkA4-/0/1771335393446?e=1785841200&v=beta&t=QXxs-aaWMfODEYGrd-J2n6Miu2rMhDXiPafPim9H8vk) personnel to enforce the law. A February [report](https://media.licdn.com/dms/document/media/v2/D4E1FAQFpc7rdRPJzlQ/feedshare-document-url-metadata-scrapper-pdf/B4EZxr0RpaIkA4-/0/1771335393446?e=1785841200&v=beta&t=QXxs-aaWMfODEYGrd-J2n6Miu2rMhDXiPafPim9H8vk) by Pour Demain estimated the unit supervising general purpose AI would need 160 staff by 2030. As of July, it had around 40, according to Farrell. Pour Demain has argued that the AI Office needs more administrative flexibility, competitive salaries and fast-tracked recruitment to attract more talent, similar to the hiring model used by the United Kingdom’s AI Security Institute.\n\n**The first enforcement tests **\n\nThe AI Office’s first enforcements are unlikely to involve blockbuster fines. More likely, it will begin with requests for information, structured dialogue with model providers and independent evaluations.\n\nUltimately, however, Brussels has the power to suspend providers from the EU [market](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-93) and issue fines that could be worth hundreds of millions. “Taking your model off the market is the ultimate escalation,” said Farrell, noting that the cost of compliance for the largest companies is tiny compared with the value of access to the European market.\n\nThe coming months will reveal exactly where the AI Office will focus its attention.\n\nCompanies that refused to sign the Code of Practice may face closer scrutiny, according to Uuk. Those include U.S.-based model providers such as [Meta](https://www.linkedin.com/posts/joel-kaplan-63905618_europe-is-heading-down-the-wrong-path-on-share-7351815758529630208-SNDY/), which declined to sign the code in its entirety, and xAI, which didn’t sign chapters on [transparency and copyright](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai). Chinese model providers also refused.\n\nBut the AI Office may watch most closely the companies with the “highest impact capabilities” that are “proving dangerous,” Farrell said. That means models from OpenAI and Anthropic that in recent months have sparked concerns about emerging cyber capabilities and loss of human control. In May, Anthropic’s [Mythos model](https://www.theparliamentmagazine.eu/news/article/anthropic-shuts-the-eu-out-of-its-most-advanced-cyber-ai-model) was pulled off market due to concerns about offensive cyber capabilities, and in July, OpenAI’s latest models autonomously [hacked the servers](https://www.politico.com/news/2026/07/28/openai-rogue-models-hugging-face-breach-01014572) of the popular AI platform Hugging Face.\n\n“It's not so much about enforcement for enforcement's sake, it's about fulfilling the goals of the AI Act with limited resources,” said Farrell. “In this case that means reducing risks around cyber and loss of control, which have both emerged recently very publicly.”\n\nEuropean developers are unlikely to fully escape scrutiny. The Future of Life’s summer [2026 AI Safety Index](https://futureoflife.org/ai-safety-index-summer-2026/) gave failing grades to the U.S.-based xAI and France’s Mistral, a signal they may be out of compliance with key sections of the AI Act.\n\nThe AI Office’s success will depend on avoiding politicization, prioritizing the largest safety risks and being adaptable to rapid technological changes Cevik said. “The AI Office has a very big task ahead to both respond to emerging needs, and also show that what they come up with is working and functioning well.”\n\nEurope also enters this new phase with more international support than it has enjoyed in the past. In the U.S., states such as Illinois, New York and California are advancing their own frontier AI rules, while Chinese President Xi Jinping said in July that [he wants his country](https://www.politico.eu/article/eu-ai-artificial-intelligence-safety-us-china/) to lead on global AI governance.\n\n“Europe should not feel entirely alone in this,” said Uuk. “The geopolitical tide is turning towards greater regulation of these frontier models.”\n\n**Sign up to *** The Parliament's* weekly newsletter\n\n*Every Friday our editorial team goes behind the headlines to offer insight and analysis on the key stories driving the EU agenda. *[Subscribe for free here](https://www.theparliamentmagazine.eu/more/newsletter-registration.htm?utm_source=footnote&utm_medium=footnote&utm_campaign=newsletter).", "url": "https://wpnews.pro/news/europe-gets-ready-to-police-frontier-ai", "canonical_source": "https://www.theparliamentmagazine.eu/news/article/europe-gets-ready-to-police-frontier-ai", "published_at": "2026-07-31 12:34:56+00:00", "updated_at": "2026-07-31 12:52:41.163528+00:00", "lang": "en", "topics": ["ai-policy", "ai-safety", "artificial-intelligence"], "entities": ["European Union", "European Commission", "Future of Life Institute", "Risto Uuk", "OpenAI", "Anthropic", "Meta", "Alphabet"], "alternates": {"html": "https://wpnews.pro/news/europe-gets-ready-to-police-frontier-ai", "markdown": "https://wpnews.pro/news/europe-gets-ready-to-police-frontier-ai.md", "text": "https://wpnews.pro/news/europe-gets-ready-to-police-frontier-ai.txt", "jsonld": "https://wpnews.pro/news/europe-gets-ready-to-police-frontier-ai.jsonld"}}