“If I go to a circus and the circus has a lion and the lion escapes and hurts me, I don't sue the lion. I sue the circus… It's not because the lion doesn't have intentionality — it does, or some sense of autonomy. It's because the corporation had a dangerous thing and knew it had a dangerous thing.”
Scaling Laws · September 2026
A student at Columbia Law asked how autonomous agents fit into the existing legal system. Bakalar, who says she is borrowing the analogy from a group chat, argues that a lot of current liability law already covers agent harms and we don't have to wait for a specialized AI statute. The open question, she adds, is whether courts will actually apply it.
Transcript
I was hoping that you could talk a little bit more about autonomous agents and possible interactions with the existing
legal system. Okay, so excellent question on autonomous agents and the intersection with our existing legal system. Any immediate take? I have one or two thoughts. I
have one or two thoughts too. We do have a legal system that could theoretically cover a lot of the things that we're concerned about. I'm borrowing this from a chat thread I'm on where we were discussing exactly this. But like if I go to a circus and the circus has a lion and the lion escapes and hurts me, well, I don't sue the lion, right? I sue the circus, that the circus is responsible for what's about to happen to me. And it's not because the lion doesn't have intentionality. It does, or some sense of autonomy. It does. It's because the corporation had some, had a dangerous thing, knew it had a dangerous thing. And whether they took the appropriate cautions or not, and that does sort of determine what kind of legal resource I might have, it's their responsibility to protect me reasonably from that thing. I think a lot of our existing law actually can be applied. And the question is sort of whether or not. It will be applied in the case of these agents. And that's going to be something that we work through in practice. There might, as things go on, I think, be need for more specialized law, of course, but we don't have to wait on that just yet. We've got a lot of tools in the toolbox.
I would agree that there are a lot of tools, and I think there's need for smart work for those of you looking for a student note to be done on how exactly some of these existing regimes would apply. The e-signature act was written, I think, in the 2000s, but don't quote me on that. But in many ways, anticipated the idea that one day we may have autonomous agents entering into legal agreements, for example, on behalf of someone. And to what extent can you then apply that back?
They already do because they can sign terms of service.
Right. So this is, you know, the confines and the application of the e-signature act, I think, is really interesting. But there are limits on some existing paradigms that we would want to apply to agents, for example. The Computer Fraud and Abuse Act, the CFAA, generally requires intent for something like a hack. And so no one at OpenAI, in theory, intended for that hack to occur, right? So that's probably not going to be a successful claim under the CFAA. And so those are the sorts of laws where we may need to see some degree of specialization. Antoine?
More from Scaling Laws
