Who's on the hook when AI goes wrong? The EU is reopening that question.

Europe scrapped its plan to write a law that decides who pays when AI causes harm. Influential members of the European Parliament are now reviving it, and OpenAI's Sam Altman is publicly backing the move, according to an exclusive interview with Decoded by POLITICO released Monday.

The proposal in question is the AI Liability Directive, a draft rule the EU shelved last year. It would have set clear rules on compensation and responsibility when AI makers and agents cause damage. Parliamentarians pushing to revive it now have a new and unexpected ally in Altman, who told POLITICO that companies should be responsible when their models cause harm during training.

The reversal is the story. Two years ago, Europe's biggest AI vendors treated liability rules as a fight to avoid. Now the CEO of the most prominent American AI lab is on record asking for the framework to come back. Inside the Parliament, lawmakers read that as political cover for a push that had stalled.

The open question is who, exactly, would be on the hook: the developer who built the model, the deployer who shipped it into a product, or the end user who triggered the harm. That choice will decide whether a liability rule reads as a deployment license for capable AI in Europe, or as a compliance moat that only the largest, best-resourced players can afford to clear.