IA : des députés européens veulent tenir les entreprises responsables si leurs modèles font des dégâts

European lawmakers are urging to broaden the scope of the EU's landmark AI regulation to hold U.S. tech giants responsible for the risks posed by their advanced models, according to written exchanges consulted by POLITICO. The proposal from four top European Parliament elected officials would add new rules to the 2024 AI Act to ensure companies like OpenAI and Anthropic can be held accountable for unintended technology use.

This shows Brussels is acknowledging the reality of AI-related damages and the growing risks, marking the first concrete attempt within the EU to address safety concerns around AI dominating global political agendas. The group calls on the Commission to present an AI responsibility law "to fill the existing legislative gaps concerning such incidents, so that no European citizen or company is left without recourse against these major providers when they cause harm," according to a note sent to the European executive on September 16th and consulted by POLITICO.

Dutch Green Party MEP Kim van Sparrentak, a member of the group, urges Europe to move faster than two years ago's first major regulatory legislation for advanced AI. She and the three other lead MEPs argue Europe must take the lead. Their call received support from European People's Party (center-right), Socialist, and Liberal Renew group members, showing broad political backing.

The threat to humanity This week marks another significant step in global AI safety efforts, following a summer with many AI-related hacking incidents and warnings about humanity's existential threat. The topic is at the center of UN General Assembly debates in New York and a G7 digital ministers meeting on Tuesday, as well as the summit this week between Chinese leader Xi Jinping and U.S. President Donald Trump.

The European Parliament's appeal highlights the AI Act's shortcomings, as its rules were drafted when AI agents pirating other companies seemed more like science fiction. AI responsibility rules could make it easier and cheaper for victims of technology-induced harm to obtain compensation, for example by shifting the burden of proof onto companies.

The new focus on product liability in Europe is also in line with some U.S. positions, including former President Donald Trump's AI adviser David Sacks' support at the POLITICO Decoded Summit in Washington last week. The European Parliament's proposal to introduce new responsibility rules marks a sharp contrast to Brussels' official stance.

For weeks, Brussels has been reacting to a series of AI incidents, researcher warnings, and tech leaders' calls to slow development with a ready-made AI Act to prevent companies from losing control of their models or agents. This renewed attention to responsibility is a tacit admission that AI legislation alone isn't enough. However, introducing new rules also poses legislative challenges.

The EU has recently revised the entire product conformity regulation for safety standards, with a general update of product liability rules set to enter into force on December 9th. Brussels also aims to ease regulation in the technology field to "simplify" rules and foster business growth. AI-specific liability rules were proposed by the European Commission in September 2022 but withdrawn the previous year.

Legal protections The current EU AI regulation already incorporates a safety angle, encouraging companies to assess their AI models or applications and determine the public risk they pose: unacceptable, high, limited, or minimal. However, the law does little to remedy the damages caused. It doesn't apply to AI models before deployment, during training or testing phases, even though most hacks last summer were conducted by internal-use-only models.

"The current system virtually recognizes no rights for people harmed by artificial intelligence," notes senior research specialist Mario Mariniello.