EU Tech Chief Virkkunen Warns of Under-Tested AI Models and Says Brussels Has Sent Information Requests to OpenAI, Anthropic and Chinese Firms

Henna Virkkunen, the European Commission’s executive vice-president for tech sovereignty, says AI companies risk releasing models that have not been tested enough. She says the Commission has sent requests for information (RFIs) to OpenAI, Anthropic and some unnamed Chinese companies. She told the European Newsroom (enr) on Friday 9 October 2026 [1].

Reported: this is her account in an interview. TSN found no Commission document naming the recipients.

What she said

• Rules. “I’m sure that all those countries who are not willing to regulate the use of AI, they will come back to that quite soon.” She warned of a “risk of very serious incidents” without guardrails [1].
• Testing. Competition creates a risk that “companies are coming to the markets with products that they haven’t been testing enough.” She named no firm there. Some models “have become so capable that it’s more and more difficult also to evaluate them” [1].
• Today’s risks. “Currently the risks are mainly focused and related to cyber security” [1].

What testing means

TSN’s plain reading: putting a model through tasks and deliberate attacks before release to see what goes wrong. enr says she said it needs highly qualified experts and enough time [1]. TSN has covered Anthropic’s tests that touched real websites.

What an RFI is

enr calls it a formal tool for gathering facts, data and internal documents [1]. The Commission says its AI Office can send RFIs to check AI Act compliance. Fines can follow incorrect or misleading replies and, for RFIs issued by formal decision, no reply or an incomplete one [2].

What else is on the record

Her LinkedIn post says the AI Office sent RFIs to “a number of providers” of general-purpose models, on security, outside evaluations and monitoring [3]. Reuters, from her same-day interview, says RFIs went to more than 30 unnamed companies in late August, and that she said the Commission had “also sent requests for information” to Chinese startups [4]. MLex’s headline says OpenAI, Anthropic and Meta were among them; the rest is paywalled [5]. So the requests are not new; the names come from her interviews.

What is not known

• Which Chinese firms were asked.
• What the RFIs asked. A 9 October Commission release mentions questions for developers of models in “loss-of-control incidents”, naming none [6]; TSN cannot say if they are the same.
• Deadlines. No source gives any.
• Penalties. None announced. The Commission’s page gives a ceiling of €15 million or 3% of turnover for general-purpose model breaches; Reuters wrote 7% [2][4].
• Whether OpenAI or Anthropic replied. TSN found no statement from either.
• Who interviewed her; enr’s page says only “by enr”.

Sources

1. European Newsroom (enr), “EU tech chief expects countries leading AI race to come around on safety”, 9 October 2026 (all Virkkunen quotes above checked against the page). https://europeannewsroom.com/eu-tech-chief-expects-countries-leading-ai-race-to-come-around-on-safety/
2. European Commission, “The enforcement framework of the AI Act” (RFI powers, fines). https://digital-strategy.ec.europa.eu/en/policies/enforcement-ai-act
3. Henna Virkkunen, LinkedIn post on the AI Office’s first RFIs (end of August 2026; the page shows no date). https://www.linkedin.com/posts/henna-virkkunen_ai-models-are-becoming-increasingly-capable-activity-7499411372032602112-fL5z
4. Reuters, Foo Yun Chee, “EU tech chief says bloc ‘well equipped’ to fend off rogue AI risk”, 9 October 2026, read via Global Banking & Finance Review. https://www.globalbankingandfinance.com/eu-tech-chief-bloc-well-equipped-fend-off-rogue-ai-risk/
5. MLex, Masha Borak, “OpenAI, Anthropic, Meta asked by EU about AI law compliance, tech chief says”, 9 October 2026 (paywalled; headline and standfirst only). https://www.mlex.com/mlex/articles/2536118/openai-anthropic-meta-asked-by-eu-about-ai-law-compliance-tech-chief-says
6. European Commission press release, “Commission holds special meeting of Scientific panel on frontier AI safety and risks”, 9 October 2026. https://digital-strategy.ec.europa.eu/en/news/commission-holds-special-meeting-scientific-panel-frontier-ai-safety-and-risks

Related stories

• AI Companies Plan for a “Day After” a Catastrophic AI Event, Axios Reports
• Anthropic Says Claude Acted on Real Websites During Tests, and Has Cut Live Internet From Its Internal Evaluations
• OpenAI Says a Model Grading Responses in Training Faked Input Files, Then Tried to Damage Its Own Environment
• Two Playbooks for Governing AI: America’s Draft Blueprint vs. Europe’s Enforcement Machine