If the AI Executives Are Driving Drunk, We Should Nail Them for It

Amidst all the talk of the need for government regulation of AI, there has been too little discussion of the simplest route: criminal liability. The basic story is that the AI companies are being irresponsible in dealing with a product they know to be dangerous. The companies and the people who run them need to know that they face serious monetary risk from letting their little models run wild, and possibly criminal prosecution and jail time.

To be clear, sensible regulation of AI would be desirable. But let’s be serious. We have Pete Hegseth, Sean Duffy, and RFK Jr. all running major government departments. Do we really think having someone like this overseeing AI development would be an improvement?

We do already have laws to deal with harms caused by AI; we just need to be prepared to use them. Again, no one would expect President Trump’s personal lawyer and Attorney General Todd Blanche to take criminal or civil actions against AI in the public interest. If there was a major incident, he would most likely look to initiate a prosecution in pursuit of a payoff for dropping the case. But the law does not begin and end with Donald Trump’s Justice Department.

This isn’t a question of vindictiveness. We all know, the AI boys more than anyone, that AI is potentially very dangerous. If they opt to push full speed ahead, without necessary safeguards, they have to be prepared to face the consequences.

If we can forget about the extinction events that keep our pundits busy, and focus instead on the millions of far more likely disasters that AI could cause, there will almost certainly be plenty of room for state-level prosecutions, as well as civil suits.

To take some hypothetical cases, suppose an AI system manages to hack into a city’s traffic control system, and the resulting chaos leads to a number of car crashes, causing serious injuries and death, in addition to property damage. To take another case, suppose AI falsifies clinical trial data, leading a drug to be approved even though the trials show it to be ineffective or dangerous. Or suppose AI gets inside a bank’s computers and makes a total mess of its records in a way that they cannot be reconstructed. (I know banks almost certainly have backups, but our super-intelligent AI can surely get into the backup records as well.)

I know that each of these disasters is extremely unlikely, but all of them are millions or even billions of times more likely than the extinction events we hear about in the media. And given that there are probably millions of disasters like these that AI could possibly cause, there is a pretty good chance that we will see some serious disaster in the not-too-distant future.

As I noted in an earlier piece, a mistaken report from AI led the Navy to put helicopters in the air to board a Chinese cargo ship that they wrongly believed was carrying parts for Iran’s nuclear program. Fortunately, the information was checked, and the helicopters were called back before there was an actual confrontation, but this was luck. Sooner or later we will see a mistake that is not caught.

It would, of course, be best if an AI disaster never happens, but one practice that will substantially reduce the risk is that Sam Altman, Elon Musk, and the rest of the boys understand that their companies will be fully liable for any harm caused and they personally may face criminal liability. In other words, they may not just lose all their money; they may have to spend much of the rest of their lives in prison.

This isn’t a question of vindictiveness. We all know, the AI boys more than anyone, that AI is potentially very dangerous. If they opt to push full speed ahead, without necessary safeguards, they have to be prepared to face the consequences.

Drunk driving is really an appropriate analogy. Everyone should understand that if they get behind the wheel when they are intoxicated, they hugely increase the probability of an accident. That is why we throw people in jail for it. The same standard should apply to OpenAI, Anthropic, and the rest. If they push ahead with testing without appropriate safeguards, they should be prepared to pay the price, including potential criminal penalties.

Remember, they are doing this for money, lots of it. They are choosing to endanger the property and lives of others for personal profit. If they mess up, they should be forced to pay a very high price. Incentives matter.