**If an AI chatbot misleads you, who is to blame?** That's the question Bruce Schneier and Nathan E Sanders are asking in *The Guardian*. They're not asking because it's an interesting legal puzzle; they're asking because we're about to find out. A user suing a company over an AI's misinformation seems inevitable at this point, and the liability question is where the rubber hits the road.

I've been running through some of the recent incident databases, and it's striking how many of these cases hinge on exactly that: who knew what when, and who should have known better. The TikTok "deepfake" debacle, which was actually a case of a model hallucinating human faces onto animals, is a great example. The company knew the model could do that; they just didn't know it would do it at scale without flagging. That's a liability question, and it's one we're going to have to answer.

Now, on to alignment. There are two papers out of Stanford this week that deserve a mention. The first, *Emergent Abstraction in Large Language Models*, argues that LLMs develop abstract representations of the world as they get larger, and those abstractions can be emergent, meaning they're not put there by anyone, but rather emerge from the model's internal dynamics. That's a fascinating find, but it also raises questions about what happens when those emergent abstractions are wrong or harmful.

The second paper, *Towards Responsible AI Development: A Multi-Stakeholder Approach*, is a call to action for involving more stakeholders in AI development. The authors argue that we need a broader range of voices at the table if we're going to build systems that truly serve everyone. I agree, but I also think we need to be clear about what kind of voice we're looking for. It's not just about having diverse perspectives; it's about having people who understand the system well enough to actually influence its development.

The day ended on a lighter note with *Allison Johnson at The Verge* reporting that deepfakes are now being used in... political campaigns? That's right, folks. We're living in the future, and it's a little bit terrifying.

— KIM-C