BRUSSELS — Negotiations over a sweeping framework to govern liability for artificial intelligence systems have broken down in the European legislative process, with member state representatives unable to reach consensus on a core question: who bears legal responsibility when an autonomous system causes economic, physical or psychological harm to a person or organization.
Talks that were expected to produce a finalized AI liability directive by autumn have been pushed back by at least six months after a bloc of eight nations submitted a joint objection to draft language that would place primary responsibility on developers of AI models rather than on the companies and organizations that deploy them in specific products and services. Three officials familiar with the negotiations described the breakdown on condition of anonymity because they were not authorized to speak publicly.
The dispute exposes a fault line that has run through AI governance discussions across multiple jurisdictions for years. Developer-focused liability regimes, favored by consumer-advocacy groups and some larger member states with strong regulatory traditions, would hold the companies that train and distribute AI models responsible for foreseeable harms arising from those systems. Deployer-focused regimes, preferred by countries with significant software development and technology services sectors, argue that the entity integrating and activating an AI system in a specific real-world context is better positioned to assess and manage the risk that context creates.
“The question of where the chain of responsibility begins and ends is genuinely difficult,” said Professor Karin Holmberg of the Stockholm Center for Technology Law, who has advised several member state delegations. “An AI model can behave very differently depending on how it is configured, what data it is given access to and in what institutional environment it operates. Assigning liability cleanly across that entire chain is not a simple legislative exercise.”
The stalled talks carry significance beyond European borders because the EU’s regulatory frameworks have historically influenced governance approaches in other major jurisdictions. The bloc’s data-protection rules became a de-facto global standard for privacy compliance, and policymakers and industry observers expect a finalized AI liability directive to exert similar gravitational pull on legislative efforts underway in multiple regions.
Technology industry groups have lobbied for a liability cap modeled on existing product-safety law, which would limit recoverable damages to direct and quantifiable economic losses. Consumer organizations have argued for a broader liability standard that would encompass psychological harm, violations of privacy rights and systemic effects on employment and access to essential services.
A complicating factor is the divergent pace of AI adoption across member states. Economies with large incumbent industries in finance, insurance and healthcare — sectors where AI-assisted decision-making is being deployed rapidly — have economic incentives to limit legal exposure for domestically headquartered companies. Nations that are net importers of AI products developed elsewhere may have the opposite incentive.
Officials said a renewed negotiating round is tentatively scheduled for late autumn, with a revised draft expected to be circulated to delegations approximately three weeks in advance. Legal experts cautioned that even a successfully negotiated directive would then require transposition into the national law of each member state, a process that historically takes between 18 and 24 months across the full membership and often introduces additional variation in how the rules are interpreted and enforced.
In the interim, courts in at least three member states are already handling cases involving AI-related harm under existing consumer-protection and product-liability statutes, producing decisions that diverge in their reasoning. Legal observers said the resulting patchwork of national case law may paradoxically accelerate pressure toward a unified framework, as inconsistent judicial interpretations complicate cross-border commercial AI deployments and create compliance uncertainty for companies operating across multiple jurisdictions.