From 9 December, software stops being sold "as is" in the EU Body: Anything shipping with an AI model in it falls under product liability, and a lot of people building with AI don't know it's coming.

SLIDE 2 - What changed Headline: The EU dropped its specific AI liability plan and used product law instead Body: Rather than write a new framework, the decision was to treat software as a product. Anything shipped after 9 December with an AI model in it is covered.

SLIDE 3 - Why that matters Headline: The standard is strict liability Body: You don't have to prove the maker was careless, only that there was a defect. That's a much lower bar than the end user licence agreements the software industry has hidden behind for decades.

SLIDE 4 - The limits Headline: Only individuals can claim, not businesses Body: It covers personal injury, damaged property and lost personal data. So if an agent causes physical harm or leaks someone's data, there's liability.

SLIDE 5 - The gap Headline: Financial loss isn't covered at all Body: An agent that sets your coat on fire creates a claim. An agent that loses your money doesn't. Given how much of the agentic AI wave is pointed squarely at financial workflows, that seems like an odd place to draw the line.

SLIDE 6 - The open question Headline: If you build on someone else's model, whose defect is it? Body: You hold the client contract. They built the model. Working out who carries the liability across that chain will come down to case law, and that takes years.

SLIDE 7 - Closer Headline: A new legal landscape, starting in December 👇.

Full episode out now, find us wherever you get your podcasts. Upside Podcast — European venture, every week.

Worth a scroll.