In a note dated 16 September, the lawmakers ask the European Commission to put forward a bill that would close regulatory gaps and would not leave EU citizens and companies without legal protection if AI systems cause harm. The initiative proposes adding separate rules on liability for unforeseen use of models to the AI Act, the EU’s 2024 AI regulation.
The signatories are Dutch Green MEP Kim van Sparrentak, German conservative Axel Voss, Italian Social Democrat Brando Benifei, and independent Irish MEP Michael McNamara. Van Sparrentak said: “Europe must step up again”.
According to the lawmakers, their appeal has already received support from colleagues in the European People’s Party, the Social Democrats, and the Renew group. In their view, this shows a broader political coalition around the idea of separately setting out liability for damage caused by advanced AI models.
EU rules currently rely mainly on product safety logic, but do not apply to AI models before deployment, during training, or testing. The authors believe that a new system could make it easier to obtain compensation, including by shifting the burden of proof to companies.
The European Commission had already proposed separate liability rules for AI in September 2022, but abandoned the project last year. Meanwhile, the updated product liability regime is due to enter into force on 9 December, and interest in a new rule aligns with Brussels’ course of “simplifying” tech regulation.
Bruegel researcher Mario Mariniello said that the current system “recognizes virtually no rights” for those harmed by AI and thereby reduces developers’ incentives to account for risk and invest in safety. Cybersecurity experts, including Marcus Hutchins, also point to recent incidents, including the breach of Hugging Face by agents developed by OpenAI, as an argument in favor of a separate liability regime.
Sources
