Imagine a self-driving car runs a red light and kills someone. The company’s defense? ‘We didn’t know it would do that.’ Under current law, that might work. But under a 19th-century legal principle called strict liability for dangerous animals, that defense evaporates. And that’s exactly what we need for AI.
You’ve probably noticed that every time an AI messes up, the company issues a press release about ‘learning’ and ‘improving.’ Meanwhile, the person whose life was ruined gets nothing. That’s because the law is designed to protect the lab, not you. The legal system treats AI as a tool, not a risk class. But what if we flipped that?
Dangerous-animal law doesn’t ask whether the animal intended harm. It doesn’t ask whether the owner was negligent. It assigns liability based on the inherent risk of the activity. If you own a tiger, you are responsible for every scratch it makes, even if you built the strongest cage. The question becomes: who profits from releasing a known risk class into society?
This is the most powerful safety incentive we have. Forget ethics committees. Forget voluntary commitments. Strict liability forces AI labs to internalize the cost of their experiments. If they know that every mistake will cost them billions, they will think twice before releasing untested models into the wild. The most dangerous idea in AI isn’t the technology. It’s the legal fiction that labs can experiment on society without paying for the damage.
Critics will say this chills innovation. They’ll argue that AI is unpredictable, so punishing labs for unforeseeable harms is unfair. But that’s exactly the point. If you’re releasing something that could cause catastrophic harm, you should be liable for every penny of damage, even if you couldn’t foresee it. The unpredictability is the risk you choose to take when you profit from the release.
Most people think the debate is about whether AI is dangerous. It’s not. The real question is: who should pay when the inevitable happens? And the answer, according to a principle that dates back to the 1800s, is clear: the owner of the dangerous animal. Strict liability doesn’t ask if you were careful. It asks if you had the audacity to release a tiger into a kindergarten.
This isn’t about slowing down AI. It’s about making sure that when AI goes wrong, the companies that profited from its release are the ones who pay, not the victims. The law is already there. We just need the courage to use it.
FAQ
Q: What is strict liability for dangerous animals?
A: It's a legal doctrine that holds the owner of an inherently dangerous animal (like a lion or tiger) fully liable for any harm it causes, regardless of the owner's care or intent. Applied to AI, it would mean AI labs are responsible for damages caused by their systems even if they couldn't have predicted the specific harm.
Q: Wouldn't this kill innovation by making AI too risky to develop?
A: It would force labs to internalize the true cost of their experiments, which might slow down reckless releases. But responsible innovation would still thrive—companies would invest more in safety, testing, and insurance. The goal is to shift the burden from victims to profiteers, not to stop progress.
Q: Isn't it unfair to hold companies liable for unforeseeable harms?
A: No. If you choose to release a potentially catastrophic technology into society, you accept the risk of unpredictability. The profit from that release comes with the responsibility for all consequences. It's the same reason we hold owners of dangerous animals liable—they chose to own the tiger.