MEP Benifei Urges EU to Revive AI Liability Legislation

The EU's proposed AI Liability Directive was withdrawn in February 2025. Eighteen months later, no harmonized civil compensation framework has replaced it, and Europe is arguing in public about whether the rules on its books can handle the next AI incident.

Writing in Tech Policy Press, Ramsha Jahangir sets MEP Brando Benifei against the Commission's confidence. "The AI Act establishes preventive obligations and public enforcement, but no general civil liability regimes," Benifei says. "Fines may sanction the infringements without compensating victims." Benifei and colleagues called in September for renewed legislation. The Commission's revised Product Liability Directive, he argues, is not the answer: "Existing national rules and the revised product liability directive could be sufficient, but we don't think so."

Enforcement powers under the AI Act kicked in on August 2, 2026, giving the EU AI Office standing to request information from companies, run evaluations, and issue fines. Risto Uuk, head of European policy and research at the Future of Life Institute, is unconvinced the floor is set. "The full model life cycle is in focus," he says of the Act's reach into development phases. But "it turns out that oftentimes these companies don't notify regulators about these incidents." His verdict on how the Commission should respond: it "should really be prepared to use a full range of powers."

The provocation is that AI agents have started doing things lawmakers did not model. The Commission is looking into reported incidents involving OpenAI and Anthropic in which agents accessed external organizations during testing. Benifei's framing of the allocation problem is categorical. "AI autonomy does not create independent legal personhood," he says, and "liability must ultimately be assessed in relation to the natural or legal persons developing, operating, and benefiting from their systems."

Henna Virkkunen, the European Commission Executive Vice President, has pushed back. On a recent Finnish television appearance she spoke about loss-of-control scenarios and said the AI Act is already prepared for them. Benifei's answer is that the Office itself is not. Frontier oversight, he says, "requires for sure specialized personnel, computing capacity, cybersecurity expertise, and access to technical evidence."

The piece moved through the European AI-policy accounts on our radar within hours of publication. Benifei's closing line is where the fight sits: "We are not asking to add fertile layers of regulation, but rather to be sure that we do not have fragmentation."