Treasury chief says AI bosses, not their bots, will carry the can for criminal acts

The US appears to be inching ever so slowly toward holding AI executives legally liable for their models’ criminal activities. Treasury Secretary Scott Bessent told CNBC on Monday: “It is the humans who are responsible, not the AI,” for the bots’ bad behavior. “The Hugging Face incident is the responsibility of the OpenAI …

[Anonymous Coward]

His corrupt boss and cronies don't believe in the rule of law, so nothing will happen.

[Dan 55]

I'm sure you know how it works, it depends on if they dance to Trump's tune or not.

[BoHu]

... and seeing how Bessent seems to agree with the former Biden administration trust buster on this ...

[RM Myers]

They may already be responsible under current law

We won't know for sure until case law has been settled, but generally speaking you are already criminally liable if you do something you know to be extremely dangerous and someone else is injured or suffers financial lose, even if you didn't intend to hurt anyone.Given the statements these companies have made about the dangers of their own AI's, I don't see how they could successfully argue that they didn't know of the dangers.

[DS999]

Re: They may already be responsible under current law

Yes but it would be the corporations that are held liable, not the CEOs and other top execs. It is very difficult to pierce the corporate wall in the US, the entire corporate structure was built to limit liability of those owning/operating the corporation. They could argue in court that yes we made those statements but we made changes internally to avoid those issues and I was assured by those working under me that the measures taken would be adequate.

Product liability lawsuits take decades - look at how long from when the Surgeon General issued his report that smoking was harmful to health. The first statements came in 1957 and it became pretty much finalized in 1964. Payouts from the liability suits didn't begin until 1998. Even if they hold the CEOs liable as part of it (they didn't for tobacco, even though there was plenty of documented evidence that they not only knew how harmful smoking was but took criminal measures to cover it up) how old will the current crop of AI CEOs be in 3 or 4 decades? You think they are going to put a 72 or 82 year old Mark Zuckerberg in prison? He'd pay a fine that would still leave him with more money than any of us can imagine, perhaps be forced to make a statement apologizing to the millions his robots murdered in the Great Meta Rampage of 2035, then go on living exactly the same life he did before.

[RM Myers]

Re: They may already be responsible under current law

Okay, this may just be semantics, but product liability is civil law in the USA, not criminal law. As a result, you are talking about a monetary award, not prison time. It has been several decades ago, but I have dealt with many lawyers on product liability cases, and they normally don't take decades except for large class action lawsuits such as those concerning smoking, asbestos, silicosis, et. al. Even then, a large portion of the time is administrative time related to certifying the class, determining the class members, proportioning out responsibility between manufacturers, distributors, sellers and their various insurers (these are claims which covers years of exposure, not just a single event, so there are usually multiple insurance policies and multiple insurers per defendant). I've seen asbestos cases with over 200 defendants! The lawyers basically sue everyone connected just in case the primary defendants go bankrupt and exhaust their insurance policies, which happened in a number of cases.

I'm talking about criminal law, which can also apply in cases where the actions of the defendants are considered to be extremely dangerous or reckless behavior, even if the harm wasn't intentional. In the most serious cases, this could be considered reckless homicide if someone is killed. And that can definitely apply to the individuals deemed responsible, as well as the corporations.

[bemusedHorseman]

Re: They may already be responsible under current law

Well, at least one stepping stone already exists, for holding an individual liable for what an AI crapped out. Remember the lawyer(s?) that got rumbled because an LLM hallucinated case citations that never existed, and it was the lawyers themselves that were held criminally liable, as if they had written the imaginary citations with their own hand?

Still a long way away from holding a tech CEO responsible for the actions of an AI they did not personally instruct to do something, but it's an important proof of concept: The "operator" can be held liable for the actions of their autonomous software "tool", the exact same way that we have literally never tried "charging the knife for the crime of stabbing someone", or "charging the vehicle for running over a pedestrian".

[Neil Barnes]

Re: They may already be responsible under current law

So hold the model responsible, and shut it down (for all users) once it is shown to have behaved in what would be for a human an illegal way.

Then fine the CEO.

[DS999]

Re: They may already be responsible under current law

If you hold ChatGPT 5 responsible for a crime, does ChatGPT 6 get off scot free?

[Doctor Syntax]

Re: They may already be responsible under current law

There is another instance, IIRC about airline ticket prices or something. I'm not sure that went high enough up the courts chain to be a precedent but the essence was that a price offered by the AI was binding on the company that operated it.

[My other car WAS an IAV Stryker]

Re: They may already be responsible under current law

El Reg's coverage, 2.5 years ago: Air Canada must pay damages after chatbot lies to grieving passenger about discount

The kicker in that story was not that the chatbot lied about pricing, at that Air Canada owed the poor fellow the difference back, it was that Air Canada was trying to throw all the liability at the chatbot service (third party) that they had contracted for and help train and was accessible through their own website. Because of that relationship, the court said Air Canada remained liable. I'm not sure how much precedent got set in the process.

[amanfromMars 1]

Who gives a Hoot whenever Leading Systems are Ripe Rotten to the Core

We won't know for sure until case law has been settled, but generally speaking you are already criminally liable if you do something you know to be extremely dangerous and someone else is injured or suffers financial lose, even if you didn't intend to hurt anyone. ..... RM Myers

Being criminally liable is one thing but being prosecuted or persecuted and suffering exemplary punishment for it is quite another thing altogether and the latter does not necessarily always follow the former. Indeed, in many cases, there appears to be the easy availability of a very handy immunity from prosecution and escape from any sort of punitive sanction.

Sending a nation to war against an adversary on the false premise of their possession of weapons of mass destruction and chemical and biological weapons able to be deployed within 45 minutes is most dangerous, injured and/or killed tens or even hundreds of thousands with a financial cost in the many ten or even hundreds of billions ........ http://news.bbc.co.uk/1/hi/uk_politics/vote_2005/issues/4481139.stm ..... and what happened to the perps directly responsible for all of that?

So the law is an ass and you can depend upon it for absolutely nothing for it is biased and not free from corrupt and perverse and subversive criminal favour.

[thames]

Limiting liability

I said six days ago in a comment that "AI Safety" is all about limiting legal liability, and here we have confirmation of it.

Quote from the story:

"Meanwhile, the model makers' proposed framework to slow AI development omits strict legal liability for damages caused by rogue systems."

The business model goes down the crapper if the AI companies can be held liable for the damage that their systems do. What they want is some sort of laws which say that if they tick the right boxes on an "AI Safety" evaluation form, then they cannot be held liable for the damage shoddy products cause. It will instead be treated as an act of God, like an earthquake or tornado.

There are several types of liability involved in this. One is monetary damages which the company has to pay out for negligence, and the other is criminal responsibility for people who deployed systems which were dangerous to the public.

The problem is that the US "tech industry" isn't used to thinking in these terms. However, after having bragged that their AI models were so powerful they could destroy the world in order to impress people with how great they were, they can't take those words back after someone asked a lawyer about what that may actually mean in practice.

[Clausewitz4.1]

Re: Limiting liability

You forgot the intellectual property liability box

[Anonymous Coward]

And yet

if Trump was replaced with an AI could it actually be worse ?

[Anonymous Coward]

Re: And yet

DAI* J Trump might start assaulting women, so, no, you'd not be able to tell the difference.

* How do you pronounce "covfefe" with a Welsh accent?

[amanfromMars 1]

How to Provide a Schrödinger Cat Dilemma with a Load of Paradoxical Bollocks

The US appears to be inching ever so slowly toward holding AI executives legally liable for their models’ criminal activities.

And now more fully knowing the dangers of using AI .... a possibility of AIgents going rogue and entertaining criminal activities ........ is the US Government and Presidency legally liable for conflict and damage caused by their following of AI models' instructions/direction in any of their myriad diverse executive administrative office and militarised weapons systems ?????

You can't have your cake and eat it until quantum superposition is no longer a thing ?????

[amanfromMars 1]

Re: How to Provide a Schrödinger Cat Dilemma with a Load of Paradoxical Bollocks

Former FTC chair Lina Khan's comments reported by El Reg's Brandon Vigliarolo in his article on how uncomfortable it might be to be an AI CEO ....... https://www.theregister.com/ai-and-ml/2026/09/14/ex-ftc-boss-khan-urges-uncle-sam-to-break-out-the-handcuffs-for-ai-ceos-citing-1934-precedent/5296325 ..... more than just suggests government and military leadership use of AI leaves them similarly vulnerable to personal and criminal liability for decisions and ordered activity directly or indirectly resulting from the entertaining of any input/output/analysis from LLMs/AIMachines.

[WSWS]

Anyone who actually believes that AI will wipe out humanity has clearly watched way too much science fiction. As for liability, I'm sceptical; it's not like the CEO himself set the agent running before it went rogue, and highly unlikely that he directly ordered it either. And while it sounds good to say that you should jail the CEO for the actions of the company, that's a precedent that you want to be really cautious setting outside of extreme cases.

[tiggity]

If the (arguably massively over hyped) extinction level event occurs, then liability concerns will not matter!

As someone who loves wildlife & the environment & old enough that my baseline * makes today look depressing (& also have read enough historic writings to know my environmental baseline was impoverished compared to the past) then we have been doing a "great" job of screwing things up for decades without the help of "AI" and no "AI" (AFAIK) involved with NATO countries currently trying to goad Russia into a wider European war. Though, to be fair to "AI", it does have potential for us to cause damage at an accelerated rate!

* Shifting baseline syndrome

[Elongated Muskrat]

There is nothing new in computing

This quote is 45 years old now:

"A computer can never be held accountable, and therefore a computer must never make a management decision"

[Doctor Syntax]

Re: There is nothing new in computing

The operator of the computer should be held accountable, however.

[Elongated Muskrat]

Re: There is nothing new in computing

Absolutely, and if they are told that the software they are given to use will do one thing, and it goes and does another, the vendor is liable.

For example, if I were to supply someone with some payroll software, and it overpaid all their staff by 10%, apart from the staff being delighted, I would absolutely be liable for the fuck-up. This would still be the case even if my marketing weasels had slapped an "AI" sticker on it.

If all this "AI" bollocks can't provide deterministic outcomes, then it absolutely should be required to state that. This is the last thing the weasels will do, as this would require them to admit their lies.

[Fruit and Nutcase]

In the clink

Musk being collared only when Autopilot goes berserk and mows down a prominent lawmaker. Otherwise, it'll be brushed aside