But if you told an embodied AI to do a home cleaning task knowing it's a possibility it will decide to pick up a gun and start spraying bullets, then society has an interest in holding you liable.

Is that the case, though? This is all pretty new tech. Did people know the machine was dangerous and didn't care, or did they not anticipate such a behavior?

It's artificial intelligence. If you treat something like a person and delegate to it, how can you later say you didn't trust it would execute your commands faithfully? And how wouldn't you be negligent in delegating a task to new, unproven tech?

Because that's how technology goes from being new to being understood. People try it, see what it can do, what it's good for, and in what ways it's dangerous, if any.

So if I make a new object that can kill you, I can try it, see what it can do, etc without liability? We already have computer programs that inflict damage on computers. Is AI new in that regard?

If companies are spending so many resources on "alignment" can they argue that they didn't know the risks of their research?

"Is that the case, though?"

Hard to make a case in 2026 that you can be oblivious to the potential of AI to do unexpected things that could impact other people.

"Did people know the machine was dangerous and didn't care, or did they not anticipate such a behavior?"

This framing of the issue as binary being either intentional negligence or blissful ignorance is not how liability works. If you are working with something potentially dangerous and impactful to society you no longer get to say "I didn't know" because we as a society hold you to a higher standard for getting the privilege of working with the potentially dangerous and impactful thing. Doctors, lawyers, engineers, pilots, truck drivers, etc. are all examples of this.

A trucker whose brakes fail causing a crash is liable even if the failure wasn't caused by intentional negligence because we hold them responsible to make sure they won't fail to begin with. A programmer and company whose safety measures fail resulting in a program hacking another company should absolutely be held accountable.

So we can have vicarious liability for banal stuff like contractors but not for AI?

You might not want to be liable, but you absolutely should be.

I might not want to be held liable if my car accelerates uncontrollably, but i am until i successfully sue the manufacturer for the defects.

Demand ai cleaners that are capable of being held accountable, or accept the risk you choose. Or just clean your own house.

Might not want to, but perhaps still should.

I don't want my dog to bite a young child, but when it does, I'm still liable, whether I want to or not.