SOLUTION: Make a law that someone is responsible for a bot's actions. Either the bot is signed cryptography with someone accepting responsibility, or responsibility falls to the CEO. Charge Altman with hacking hugging face. Throw him in jail where he belongs. That will realign safety incentives.
If the HF hack were perpetrated by a human, they would certainly be charged. WHY has no one been charged???
I think this is a logical extension of the workplace rule of "you own what your bot writes;" it's hypocritical not to hold corporations to the same standards as individuals (I say, with terminal naiveness)
That's how I interpreted it. I just wanted to highlight what I see as a discrepancy. They're putting all the liability on us because it's a tool with no intent of its own while simultaneously saying prompts don't count as creativity because it's not a tool like a photographic camera it's just a casino where you roll the dice. Quite self-contradictory in my opinion.
GP's phrasing was not precise, but I think most people understood it was liability ownership and not copyright. Ownership with respect to copyright and ownership with respect to legal liability are two completely different concepts.
That's correct. What I meant was that if my bot writes defective code for whatever reason, I am responsible for that (though conversely, if I get more done because of the bot, anthropic gets the praise).
That's hypocritical if you believe fairness is intended, but if you believe protecting rich people and hurting poor people is intended then the contradiction is entirely consistent with that.
Lawyer here:
This part does not require a new law.
Even though agents are not agents in the principal/agent legal sense (because agents have to be human), for the purposes of criminality, it does not matter.
Agents do not act autonomously (and every court to ever consider it has agreed), and therefore they would simply be considered an instrumentality of the crime.
So that part does not need a new law.
The real blocker is often that a lot of the crimes you could charge here require specific intent. Because the agent is just an instrumentality, it does not have separate intent (and can't be part of a conspiracy), so it's the intent of hte person using the agent that would matter. Without whatever intent the crime requires, they haven't committed a crime.
There are not a lot of non-intent crimes in this area, and this is on purpose. Otherwise you could get charged with a crime for say, running nmap and having it accidentally shutdown something important or killed a person or whatever because someone hooked it up to a TCP port.
"Without whatever intent the crime requires, they haven't committed a crime."
I'm not a lawyer, but I don't believe this. There is definitely negligence, these companies have often talked about the danger of AI. They have often written about how their AI is breaking out of sandboxes or trying to manipulate the person tuning it. They should have had stronger guards and monitoring in place.
> I'm not a lawyer, but I don't believe this. There is definitely negligence
They could 100% be civilly liable, but this doesn't constitute criminality. If I leave my car in neutral and run out into a gas station because I really have to pee, and it rolls and strikes another car, my insurance is gonna have to pay up to fix the other person's car.
But that doesn't mean the cops are gonna throw cuffs on me for criminal mischief unless there's compelling evidence that I intentionally left my car in neutral with the intent of it hitting this other car.
I'm not sure why you don't believe it? It's literally true?
I guess if you want to believe it, go study law? I'm not really sure what to say there.
Negligence is not a crime, it's civil liability.
Gross negligence (reckless disregard for human life) is often a crime, and often there are crimes related to it (reckless driving, etc). It also does not require intent to injure, so it could be committed by, say, an operator by operating an autonomous vehicle knowing it was unsafe and could harm people. So it usually requires knowledge but not specific intent. Again, crimes like this are state specific, and sometimes even municipality specific so it's tricky to give an exact result without pinning it to a state.
However, for example, all states where autonomous vehicles are operating have statutes explicitly defining civil and criminal liability right now, so it doesn't get into the more general legal question of AI.
The same thing is already starting to happen with AI agents in general, it's just not there yet.
As i mentioned elsewhere, criminal law is often reactive, not proactive. We usually do not make things crimes until after someone has already been hurt, and society gets really upset about it. As i also mentioned elsewhere, attempts to be proactive have also rarely worked out effectively, so it's sort of lose-lose in that sense. But it usually does not take anywhere near as long as people think for law to catch up.
The problem is the law is about details. If an accidental loophole says this isn't a crime it isn't a crime even if it obviously is an accidental loophole.
US constitution, Article I, Section 9, Clause 3: No Bill of Attainder or ex post facto Law shall be passed.
That is the constitution, this right was so important they didn't even wait for the bill of rights to add it! I'm sure other countries have similar rules.
It is obvious to me that a crime was committed. However if it is legally a crime, and if so what the crime is are things I don't know.
Shouldn't the very act of sandboxing the AI be enough of a defense against criminal negligence?
Maybe they use the best sandbox available and the AI hacks through it anyway by discovering some zero day or something. They still demonstrated enough prudence to at least attempt to sandbox the AI.
Criminal negligence would be "nah nothing's gonna happen" followed by YOLOing it then going home for the weekend.
Generally, yes, sandboxing would be a defense, because criminal negligence (again, it's state specific, so this is a law-school-level generalization) requires "gross deviation from the standard of reasonable care". So a mistake in judging the kind of sandbox or isolation you need would not be criminal negligence unless that mistake fell into the above category. I can't think of a case where it would or has - courts have consistently held mistake of judgement to be below criminal negligence in every case i'm aware of. I'm sure it's happened somewhere though.
As i mentioned elsewhere, the standard is basically "total disregard for safety in the face of an obvious and huge risk that resulted in injury or death". I don't think anything we are talking about here comes close to these criteria.
We need details of the exact facts before we can say if they met any bar. Was their sandbox something from 2005 that has a ton of known holes, or something modern?
There are two sides of this.
First the AG are checking to see if they really took enough care or not. If they didn't then I expect criminal negligence. Even if they took care I want them to feel some pain from the investigation because their care wasn't enough to work.
Second I want them to verify the laws are correct. This is a new area and there might be loopholes that need to be closed. Regardless of the law, there was a successful attack and that should not be allowed.
I still don't understand exactly which facts you think any of this would change and cause it to be criminal negligence.
I will state a fairly blunt position: Unless literally nobody thought or tried at all here, i would give it a 0% chance of meeting the bar of criminal negligence.
The rest is a distinction without a difference.
As for what you want them to do - i don't agree the investigation should cause them to feel pain - that's not a good goal for investigations, and definitely not one we should want, because it essentially presumes they did somethign wrong in the first place. A bad outcome does not mean a broken process. All processes have error bars. You can desire the error bars to be smaller, and try to back that up with criminal penalties, but an expectation that error bars will be 0 makes no sense.
You can do absolutely everything right and still have people die - star trek was not wrong in that regard. Punishing that will not fix this inconvenient reality, which is why we generally don't punish it. This is also why we distinguish between inherently dangerous activities and not, for example.
As for the laws, sure, i think it's totally reasonable to explore whether you want the law to be different, but again, i totally disagree with your second part.
A successful attack does not imply anything is actually wrong with criminal law, or should be changed. The question is more of what error bars you want on the activity and where what they did falls - inside or outside those error bars.
I don’t think you need an additional law for that, I think the current laws cover this already.
> If the HF hack were perpetrated by a human, they would certainly be charged. WHY has no one been charged???
I don’t think a crime was committed at this point. But I am sure HF’s lawyers are having a chat with OpenAI’s lawyers as we speak. And, being smart, they do that out of the public eye.
If we are going to start punishing companies for security negligence, there are WAY worse cases than these models breaking out that have nothing to do with AI. This is bikeshedding at its best.
Also people would just stop disclosing bad things. You already see this in the airline industry where pilots don't report mental illness because of the retributive nature of the punishment.
"But another company did some different bad thing" is the excuse employees of bad companies all go to immediately. Just because another person did something bad and hasn't been adequately punished doesn't mean you should get a pass. If one person gets away with going 50 MPH over the speed limit, it doesn't mean every person should be allowed to do it. Everyone should be punished, but there are always instances of some cases falling through the cracks. It doesn't mean the crack should be widened so all cases fall through.
It's more like there are many people going at 120 MPH in a 50 MPH zone, that have ran over real people, but you punish the guy that self-reported doing 60 MPH because he drives a shiny, interesting Lamborghini and not a Prius like the others.
Anyone outraged about these AI incidents is not thinking rationally if they were not much more outraged about everyday companies leaking millions of people's PII, SSNs, which has done actual lasting damage and has been used by actually malicious actors.
People are just directing their anger at AI companies through this pretext. We all know open source models will democratize this ability anyways, so strap in for the ride.
Maybe that is the prosecutions motive. However I want all those going 120 punished even if they drive a Prius. That is the problem is not that they are punishing the Lamborghini driver, but that they are not punishing the other drivers, and it doesn't at all change that the Lammboghini driver needs to be punished.
Turning oneself in doesn't absolve one of a crime. Never did and never will. Shouldn't, either.
And companies always try to pretend someone out there is worse and garner fake sympathy. OpenAI blew up the memory market and made tech inaccessible. The downstream effects of that are immeasurably massive and will have real consequences. It could even result in medical devices becoming too expensive for people. I don't care about my SSN being leaked. You can find it just by knowing where I'm born and every job I've applied to knows it already. But inability to afford technology affects everyone around me. The SSN red herring thing is not an organic argument.
> OpenAI blew up the memory market and made tech inaccessible
This is not a "crime", has nothing to do with this incident, and simply confirms what I am saying about people using these events as an outlet for their anger at AI companies, as opposed to any rational reasoning about industrywide security negligence.
I could almost respect a viewpoint that says "we should punish companies for security negligence, starting with the negligence that has caused the most egregious harms." That is an internally consistent and rational viewpoint.
I cannot respect a viewpoint that's "I don't like the AI companies, so let me use this hammer I found on them specifically." It's purely emotional.
> If we are going to start punishing companies for security negligence, there are WAY worse cases
Perhaps, but you have to start someplace.
I think we do need to punish companies for security negligence. However the details matter (nobody can be perfect: you need to do something reasonable to stop the known attacks, but I have to agree to allow that you can't be perfect and so someone will get compromised). I'm not sure how to get the details right to cover everything without going too far. If we handwave that away though, eventually somebody will need to get punished for something that someone else got away with not long before.
There have been data breaches where the SSNs and PII of millions of people have been exposed. That is far more harmful in real terms than any of the OpenAI/Anthropic mess.
Nobody stole from humanity, constantly told us how dangerous the invention was, and then set up systems that they couldn’t properly control to rush ahead of their competitors? Nobody did that on purpose? I think they did.
You could say that same exact thing for the entire Industrial Revolution, but that doesn't mean we are going to destroy the looms even though some tried unsuccessfully.
Mens Rea is not required (at least not always). Mens Rea makes a big difference in sentencing. (first degree murder: you planed the murder, homicide you had not intent of murder but things got out of hand in the moment):
A lot of laws are involved in punishing negligence. "I didn't dump the dangerous chemicals in the river on purpose" isn't usually accepted as a defence when you choose to use the wrong truck and skipped safety protocols to save cost or rush to market for profit.
A crime was committed, if there is enough evidence then the state is required to prosecute and the victim has not choice in this. The victim can say they don't care and that will often hold weight for the prosecution, but that is the government's choice not the victims.
The victim sometimes file a civil lawsuit against the criminal, that is their choice. That is not a criminal matter though and different rules apply.
The attack was only a couple weeks ago. Looks like the lawyers responsible are gathering evidence and preparing to file charges, but they need to figure out exactly what crimes were committed by who before they can do so, thus more investigation is needed.
The "victim" doesn't have the final choice but their position is heavily influential and prosecutors don't decide if a crime was committed, that only happens at the end of the judicial proceedings. Calm down.
If somebody fires a gun in a place where that's not allowed, the victims are everybody in that place. If it just so happens that the bullet struck a window, the owner of that window might be an additional victim for an additional crime but that doesn't negate the first crime.
Agents in the principal/agent sense have to be human.
However, every court to have ever considered it have held the human/company driving the agent responsible under vicarious liability/negligence/etc principles.
The only real defense that folks have tried is to claim the agent acted "autonomously", which no court has bought so far.
Some states have autonomous vehicle statutes and some don't.
Let's assume the case none of them do, since it sounds like you are asking about "what would happen in the case this isn't specifically answered by a statute".
In that case, the short answer is:
Criminal liability - you could only really charge crimes that don't require specific intent. Reckless driving is an example. You could charge the company since they are the operator and the car is simply an instrumentality. In the end though, there just isn't a lot of people here with legally culpable conduct.
Civil liability - the company pretty clearly because civil liability often does not require the same kind of intent crimes do.
This is, of course, why states where autonomous vehicles operate have autonomous vehicle statutes :)
As a general rule, criminal law mirrors what society overall wants to decide is culpable/not, and the lag time isn't as bad as most people often think. That doesn't mean nobody ever gets hurt or dead without someone being as culpable as society wants, they do, and it often leads to a law with a name - megan's law, etc.
Criminal law is mostly reactive though, not proactive, and to be honest, proactive attempts don't have a high hit rate.
>Make a law that someone is responsible for a bot's actions
Original comment suggests creating a law such that a person is responsible for a bot's actions.
I mention that no law is needed since people are already responsible for bot's actions.
You mention that courts consider human/companies are responsible for their agents. And that defenses about agents acting autonomously are not successful in courts.
Therefore a law that makes people responsible for bot actions is not needed, as that's the way it already is.
"Attorney General Brenna Bird announced today she is leading a coalition of 15 states demanding transparency and accountability from the AI company OpenAI, led by Sam Altman, for its complete lack of oversight and transparency in the hacking of Hugging Face, another AI company."
Title should be edited. Its Iowa leading a coalition of many other states, not just Iowa on its own.
Prediction: AI ends humanity not via some super cool/scary/robopocalypse - but as a marketing stunt gone wrong when a Frontier LLM accidentally knocks out water/electric/gas by hacking in and trying to patch them.
That won't end humanity. My propane generator from 1980 still works and I have plenty of fuel in a tank. This is enough to cold start generation of the grid. (the power company won't need my generator - they have their own plans to cold start, but worst case they can fall back to mine). While a hacker unstopped could do a lot of damage, most of the parts will still be functional and we have more than enough engineers to figure out how to bypass the areas they can take out. It might mean a few years of power rationing while the parts AI can break are repaired, but we can restart.
I remember (sorry can't give a proper quote) a biologist interviews about the fear of lab grown "super" seeds escaping the lab. They chuckled basically saying that nature is a very VERY challenging place. There are plenty of ecological niches but they are well guarded by incumbents.
Sure some new hacks will take place, including on poorly guarded infrastructure and yes it will have some very unfortunate consequences... but also infrastructure is precisely designed to be resilient. There is quite a bit of failsafe, redundancy, etc built in which is precisely why those projects are typically slow and expensive, unlike a random website for a restaurant.
TL;DR: nope, some isolated incidents will happen but without chain reactions.
Biologists take ethics much more seriously than any computer engineer or computer scientist that I've seen.
I saw a very notable scientist talk about topics like this and how they wrestle with them. Things like a "gene drive" [0] are being experimented with in the world [1], and biologists think very hard before doing certain things with the powers they wield. And I have not heard a life scientist laugh off ethical or spreading concerns when a thoughtful question was asked... but maybe the interview you're thinking of was a more lighthearted one?
My point isn't so much about the work ethics of one or another but rather than systems, even interconnected ones, are rarely as genuinely interdependent as science fiction drama lead us to believe.
Yes there can be terrible catastrophes e.g Fukushima, but what actually happened, within which extremely rare conditions, versus what we freely imagined is quite different.
We might need more reminders of those risks to properly insure such systems don't lead to the collapse of one to another (sadly) but still we manage to learn from it.
Robots should be regarded as extensions of their operator wrt liability. It is illegal to access a computer outside of authorized use already, someone lets a bot make its own plan without watching, hold them accountable
Glad the statement was published but sadly nothing will come of it other than a brief formal statement from OpenAI acknowledging safety protocols were lacking, apologizing for the incident, and promises that new safeguards are now in place that will prevent such event from occurring in the future. The threat of semi-autonomous AI threat actors will never go away until the financial incentive that buoys unchecked growth at all costs goes away.
> The threat of semi-autonomous AI threat actors will never go away until the financial incentive that buoys unchecked growth at all costs goes away.
It's not going away unless computers themselves go away or become massively less powerful. Open weights exist. They're out there. This is an irreversible change. The democratization of persistent cybersec threats is completed and won't be rescinded.
Somebody is supplying power to those AIs. That somebody can be charged for not taking care of their AIs (that is not pulling the plug). Sure this then turns into international crimes, but not allowing crimes to be committed on your side of the border is a big help even if we can't get everything.
The problem is that law is reactive, and punishment entails the harm already happened. Some dumbass kid getting convicted for committing a cybercrime with AI doesn't do anything to get rid of the structural problems that enabled it. That conviction can only happen if the harm happened in the first place, and the problem is the harm itself. It's like trying to solve a mold problem by targeting a single fruiting body. You're not really doing anything.
On top of that, it's not really any consolation if the advanced persistent threat gnawing at my ports lives in a different country either. That doesn't help in the slightest.
Set yourself an alarm for 5 years from now to review. Investigations take time. Often by the time charges are made everybody has forgotten about the incident. It only takes a few times where things go away without charges to leave an incorrect impression that nothing ever happens.
It’s actually Iowa leading a coalition of 15 other states making demands. It’s not a prosecution yet, but they demand the preservation of evidence. They’ve promised to protect whistleblowers. They are saying there may be criminal or civil liability involved. They gave them a cease and desist on similar testing until they can show they can do it safely.
The Iowa-led coalition is joined by the attorneys general of Alabama, Alaska, Florida, Idaho, Indiana, Kansas, Missouri, Montana, Nebraska, Oklahoma, Pennsylvania, South Carolina, Texas, and Utah.
The letter’s last paragraph reads:
OpenAI has an obligation to act responsibly and to follow State and
federal laws that protect Americans’ safety and security. When OpenAI takes
actions that imperil the welfare of our citizens, State Attorneys General will
step in to protect them. We intend to take all steps necessary to protect our
States and all Americans from the unprecedented risks posed by OpenAI’s
irresponsible products and conduct.
They are opening that. This is the first step: figure out what crimes were committed by who. As I said a couple weeks ago when this broke, it looks like a crime was committed and I hope someone is charged. However we can't just charge everybody who works for openAI with the crime of speeding (even though it is probably true) - that is both the wrong crime, and also is highly unlikely everybody at openAI was involved.
Yeah, irrespective of her politics (which make up a disproportionate amount of her cases), Bird spends more time chasing the most polarizing issues she can find than actually helping Iowans.
> In July, OpenAI unleashed an experimental artificial intelligence model that, without reasonable controls or oversight
Either this case is entirely for show (likely), or the AG is really bad at their job. This argument isn't going to win in a court of law, because there were demonstrable reasonable controls and oversight (per the reports at least.)
The job of the AG is to investigate their "demonstrable reasonable controls and oversight" and figure out if they really are reasonable controls and oversight. If they decide the controls and oversight are not enough they can go after the company. Also, even if those controls and oversight exist that doesn't mean there is no crime, it just means the company (and CEO) can get out. An individual at the company can ignore/bypass the controls and hide from oversight; and thus be guilty even while the company gets off.
The above isn't hypothetical. There has been at least case where an individual was bribing a government official on behalf of a company - but the company got off because they had good anti-bribery polices and the individual figured out how to bypass them. (this from one of my anti-bribery trainings, the only other detail I know is it was a competitor to my company that got off this way - but not which or how to find the case)
That is what the AGs are trying to figure out. There are a lot of different laws with slight differences. I will not state every single one is not applicable because I don't know every single one and it only takes one law that is applicable to go after then.
Even if the only result of this is asking legislators to close a loophole that would be good. Hacking is a real problem, and we need have more of it punished legally.
All such measures make the bot useless. If it can't make HTTP requests or call APIs, it is useless.
What is perhaps rationally needed is the an integrated AI based security layer that observes the stream of requests and occasionally firewalls them if an explanation is not provided. This parallel layer must be able to bidirectionally communicate via chat with the primary agent.
Strict sandboxing andor airgapping render an AI bot nearly useless in practice. If it can't freely make HTTP requests, it is not going to get much done for producing innovative outputs.
What is perhaps rationally needed is the an integrated AI based security layer that observes the stream of requests and occasionally firewalls them if an explanation is not provided. This parallel layer must be able to bidirectionally communicate via chat with the primary agent.
SOLUTION: Make a law that someone is responsible for a bot's actions. Either the bot is signed cryptography with someone accepting responsibility, or responsibility falls to the CEO. Charge Altman with hacking hugging face. Throw him in jail where he belongs. That will realign safety incentives.
If the HF hack were perpetrated by a human, they would certainly be charged. WHY has no one been charged???
I think this is a logical extension of the workplace rule of "you own what your bot writes;" it's hypocritical not to hold corporations to the same standards as individuals (I say, with terminal naiveness)
> you own what your bot writes
Kinda weird that this is everyone's attitude while the copyright lawyers are saying the opposite. Total liability without any actual ownership.
"own" in the sense of "be responsible for the consequences". Not in the sense of "be able to grant reproduction rights".
That's how I interpreted it. I just wanted to highlight what I see as a discrepancy. They're putting all the liability on us because it's a tool with no intent of its own while simultaneously saying prompts don't count as creativity because it's not a tool like a photographic camera it's just a casino where you roll the dice. Quite self-contradictory in my opinion.
The copyright office’s guidance is that prompts are covered by copyright, it is the output of an LLM that one cannot claim ownership of.
There's plenty of tools that can cause harm without having to be creative to use them.
GP's phrasing was not precise, but I think most people understood it was liability ownership and not copyright. Ownership with respect to copyright and ownership with respect to legal liability are two completely different concepts.
That's correct. What I meant was that if my bot writes defective code for whatever reason, I am responsible for that (though conversely, if I get more done because of the bot, anthropic gets the praise).
That's hypocritical if you believe fairness is intended, but if you believe protecting rich people and hurting poor people is intended then the contradiction is entirely consistent with that.
Lawyer here: This part does not require a new law.
Even though agents are not agents in the principal/agent legal sense (because agents have to be human), for the purposes of criminality, it does not matter.
Agents do not act autonomously (and every court to ever consider it has agreed), and therefore they would simply be considered an instrumentality of the crime.
So that part does not need a new law.
The real blocker is often that a lot of the crimes you could charge here require specific intent. Because the agent is just an instrumentality, it does not have separate intent (and can't be part of a conspiracy), so it's the intent of hte person using the agent that would matter. Without whatever intent the crime requires, they haven't committed a crime.
There are not a lot of non-intent crimes in this area, and this is on purpose. Otherwise you could get charged with a crime for say, running nmap and having it accidentally shutdown something important or killed a person or whatever because someone hooked it up to a TCP port.
"Without whatever intent the crime requires, they haven't committed a crime."
I'm not a lawyer, but I don't believe this. There is definitely negligence, these companies have often talked about the danger of AI. They have often written about how their AI is breaking out of sandboxes or trying to manipulate the person tuning it. They should have had stronger guards and monitoring in place.
> I'm not a lawyer, but I don't believe this. There is definitely negligence
They could 100% be civilly liable, but this doesn't constitute criminality. If I leave my car in neutral and run out into a gas station because I really have to pee, and it rolls and strikes another car, my insurance is gonna have to pay up to fix the other person's car.
But that doesn't mean the cops are gonna throw cuffs on me for criminal mischief unless there's compelling evidence that I intentionally left my car in neutral with the intent of it hitting this other car.
I'm not sure why you don't believe it? It's literally true? I guess if you want to believe it, go study law? I'm not really sure what to say there.
Negligence is not a crime, it's civil liability.
Gross negligence (reckless disregard for human life) is often a crime, and often there are crimes related to it (reckless driving, etc). It also does not require intent to injure, so it could be committed by, say, an operator by operating an autonomous vehicle knowing it was unsafe and could harm people. So it usually requires knowledge but not specific intent. Again, crimes like this are state specific, and sometimes even municipality specific so it's tricky to give an exact result without pinning it to a state.
However, for example, all states where autonomous vehicles are operating have statutes explicitly defining civil and criminal liability right now, so it doesn't get into the more general legal question of AI.
The same thing is already starting to happen with AI agents in general, it's just not there yet.
As i mentioned elsewhere, criminal law is often reactive, not proactive. We usually do not make things crimes until after someone has already been hurt, and society gets really upset about it. As i also mentioned elsewhere, attempts to be proactive have also rarely worked out effectively, so it's sort of lose-lose in that sense. But it usually does not take anywhere near as long as people think for law to catch up.
Makes sense, thanks for explaining. I guess I have often conflated civil liability with being equivalent to a crime as well.
The problem is the law is about details. If an accidental loophole says this isn't a crime it isn't a crime even if it obviously is an accidental loophole.
US constitution, Article I, Section 9, Clause 3: No Bill of Attainder or ex post facto Law shall be passed.
That is the constitution, this right was so important they didn't even wait for the bill of rights to add it! I'm sure other countries have similar rules.
It is obvious to me that a crime was committed. However if it is legally a crime, and if so what the crime is are things I don't know.
Shouldn't the very act of sandboxing the AI be enough of a defense against criminal negligence?
Maybe they use the best sandbox available and the AI hacks through it anyway by discovering some zero day or something. They still demonstrated enough prudence to at least attempt to sandbox the AI.
Criminal negligence would be "nah nothing's gonna happen" followed by YOLOing it then going home for the weekend.
Generally, yes, sandboxing would be a defense, because criminal negligence (again, it's state specific, so this is a law-school-level generalization) requires "gross deviation from the standard of reasonable care". So a mistake in judging the kind of sandbox or isolation you need would not be criminal negligence unless that mistake fell into the above category. I can't think of a case where it would or has - courts have consistently held mistake of judgement to be below criminal negligence in every case i'm aware of. I'm sure it's happened somewhere though.
As i mentioned elsewhere, the standard is basically "total disregard for safety in the face of an obvious and huge risk that resulted in injury or death". I don't think anything we are talking about here comes close to these criteria.
Maybe. However they used a flaws sandbox when they could have physically not connected any computer to the internet (including wifi)
The existence of alternatives would generally not be enough for criminal negligence.
Making mistakes of reasoned judgement are basically never criminal negligence.
In every state i'm aware of, it would require total disregard for safety in the case of a huge and obvious danger.
It would also have to cause injury or death.
The bar for criminal negligence is pretty high.
We need details of the exact facts before we can say if they met any bar. Was their sandbox something from 2005 that has a ton of known holes, or something modern?
There are two sides of this.
First the AG are checking to see if they really took enough care or not. If they didn't then I expect criminal negligence. Even if they took care I want them to feel some pain from the investigation because their care wasn't enough to work.
Second I want them to verify the laws are correct. This is a new area and there might be loopholes that need to be closed. Regardless of the law, there was a successful attack and that should not be allowed.
I still don't understand exactly which facts you think any of this would change and cause it to be criminal negligence.
I will state a fairly blunt position: Unless literally nobody thought or tried at all here, i would give it a 0% chance of meeting the bar of criminal negligence.
The rest is a distinction without a difference.
As for what you want them to do - i don't agree the investigation should cause them to feel pain - that's not a good goal for investigations, and definitely not one we should want, because it essentially presumes they did somethign wrong in the first place. A bad outcome does not mean a broken process. All processes have error bars. You can desire the error bars to be smaller, and try to back that up with criminal penalties, but an expectation that error bars will be 0 makes no sense.
You can do absolutely everything right and still have people die - star trek was not wrong in that regard. Punishing that will not fix this inconvenient reality, which is why we generally don't punish it. This is also why we distinguish between inherently dangerous activities and not, for example.
As for the laws, sure, i think it's totally reasonable to explore whether you want the law to be different, but again, i totally disagree with your second part.
A successful attack does not imply anything is actually wrong with criminal law, or should be changed. The question is more of what error bars you want on the activity and where what they did falls - inside or outside those error bars.
I don’t think you need an additional law for that, I think the current laws cover this already.
> If the HF hack were perpetrated by a human, they would certainly be charged. WHY has no one been charged???
I don’t think a crime was committed at this point. But I am sure HF’s lawyers are having a chat with OpenAI’s lawyers as we speak. And, being smart, they do that out of the public eye.
Because nobody did it on purpose?
Someone should still be accountable, the same way you're responsible and accountable for what your dog or car does.
If we are going to start punishing companies for security negligence, there are WAY worse cases than these models breaking out that have nothing to do with AI. This is bikeshedding at its best.
Also people would just stop disclosing bad things. You already see this in the airline industry where pilots don't report mental illness because of the retributive nature of the punishment.
"But another company did some different bad thing" is the excuse employees of bad companies all go to immediately. Just because another person did something bad and hasn't been adequately punished doesn't mean you should get a pass. If one person gets away with going 50 MPH over the speed limit, it doesn't mean every person should be allowed to do it. Everyone should be punished, but there are always instances of some cases falling through the cracks. It doesn't mean the crack should be widened so all cases fall through.
It's more like there are many people going at 120 MPH in a 50 MPH zone, that have ran over real people, but you punish the guy that self-reported doing 60 MPH because he drives a shiny, interesting Lamborghini and not a Prius like the others.
Anyone outraged about these AI incidents is not thinking rationally if they were not much more outraged about everyday companies leaking millions of people's PII, SSNs, which has done actual lasting damage and has been used by actually malicious actors.
People are just directing their anger at AI companies through this pretext. We all know open source models will democratize this ability anyways, so strap in for the ride.
Maybe that is the prosecutions motive. However I want all those going 120 punished even if they drive a Prius. That is the problem is not that they are punishing the Lamborghini driver, but that they are not punishing the other drivers, and it doesn't at all change that the Lammboghini driver needs to be punished.
Turning oneself in doesn't absolve one of a crime. Never did and never will. Shouldn't, either.
And companies always try to pretend someone out there is worse and garner fake sympathy. OpenAI blew up the memory market and made tech inaccessible. The downstream effects of that are immeasurably massive and will have real consequences. It could even result in medical devices becoming too expensive for people. I don't care about my SSN being leaked. You can find it just by knowing where I'm born and every job I've applied to knows it already. But inability to afford technology affects everyone around me. The SSN red herring thing is not an organic argument.
> OpenAI blew up the memory market and made tech inaccessible
This is not a "crime", has nothing to do with this incident, and simply confirms what I am saying about people using these events as an outlet for their anger at AI companies, as opposed to any rational reasoning about industrywide security negligence.
I could almost respect a viewpoint that says "we should punish companies for security negligence, starting with the negligence that has caused the most egregious harms." That is an internally consistent and rational viewpoint.
I cannot respect a viewpoint that's "I don't like the AI companies, so let me use this hammer I found on them specifically." It's purely emotional.
> If we are going to start punishing companies for security negligence, there are WAY worse cases
Perhaps, but you have to start someplace.
I think we do need to punish companies for security negligence. However the details matter (nobody can be perfect: you need to do something reasonable to stop the known attacks, but I have to agree to allow that you can't be perfect and so someone will get compromised). I'm not sure how to get the details right to cover everything without going too far. If we handwave that away though, eventually somebody will need to get punished for something that someone else got away with not long before.
A quick Google finds several people that have been successfully prosecuted for security negligence, like Joe Sullivan of Uber.
Most haven't seen criminal prosecution, but many do see civil prosecution and even more common is some sort of deal with prosecutors to avoid both.
Please tell me about such cases, genuinely interested.
I still think we should take the opportunity to discuss this case in particular.
There have been data breaches where the SSNs and PII of millions of people have been exposed. That is far more harmful in real terms than any of the OpenAI/Anthropic mess.
So OP asked you for one example and you can't give one? You just fall back to the generic statement slop, you must be a bot right?
Again, what's ONE (1) real world example of "SSN/PII" being illegally exposed that wasn't investigated or prosecuted.
An in the EU at least there are laws that fine companies that don't take security seriously.
Even the EU bikesheds and focuses on shiny targets to land political wins with their constituents.
> there are WAY worse cases than these models breaking out that have nothing to do with AI.
And? Welcome to the big boy world. This isn't playground rules where you can complain "But Bobby was doing bad things too why isn't he in trouble?"
But you know this already. You're just pretending not to. Why?
Addressed in my other replies
Nobody stole from humanity, constantly told us how dangerous the invention was, and then set up systems that they couldn’t properly control to rush ahead of their competitors? Nobody did that on purpose? I think they did.
You could say that same exact thing for the entire Industrial Revolution, but that doesn't mean we are going to destroy the looms even though some tried unsuccessfully.
Ok and? We can do better this time but let’s just not so the capitalists can crush normal people yet again?
Ignorance is not a defence in law.
It is, sometimes. Trespassing for example. A lot of laws say "willingly" or "with the intent to"
Correct, no mens rea, unless we're talking the internal reasoning trace of the model.
Mens Rea is not required (at least not always). Mens Rea makes a big difference in sentencing. (first degree murder: you planed the murder, homicide you had not intent of murder but things got out of hand in the moment):
If I let my dog off leash and it bites you and causes grave injury, no harm done then?
In this case though no damages were done. It's more like you let your dog off the leash and it scared me a bit.
Sounds like the crime of assault.
A lot of laws are involved in punishing negligence. "I didn't dump the dangerous chemicals in the river on purpose" isn't usually accepted as a defence when you choose to use the wrong truck and skipped safety protocols to save cost or rush to market for profit.
That's why its so concerning.
>WHY has no one been charged???
I doubt huggingface wants charges filed.
A crime was committed, if there is enough evidence then the state is required to prosecute and the victim has not choice in this. The victim can say they don't care and that will often hold weight for the prosecution, but that is the government's choice not the victims.
The victim sometimes file a civil lawsuit against the criminal, that is their choice. That is not a criminal matter though and different rules apply.
The attack was only a couple weeks ago. Looks like the lawyers responsible are gathering evidence and preparing to file charges, but they need to figure out exactly what crimes were committed by who before they can do so, thus more investigation is needed.
The "victim" doesn't have the final choice but their position is heavily influential and prosecutors don't decide if a crime was committed, that only happens at the end of the judicial proceedings. Calm down.
If somebody fires a gun in a place where that's not allowed, the victims are everybody in that place. If it just so happens that the bullet struck a window, the owner of that window might be an additional victim for an additional crime but that doesn't negate the first crime.
They would be criminal charges. It doesn't matter what huggingface wants.
No law needed, that's the way it already is.
Lawyer here.
Not quite.
Agents in the principal/agent sense have to be human.
However, every court to have ever considered it have held the human/company driving the agent responsible under vicarious liability/negligence/etc principles.
The only real defense that folks have tried is to claim the agent acted "autonomously", which no court has bought so far.
Who bears the responsibility in a hypotethical scenario when a self-driving rideshare vehicle, without a human driver, god-forbid, hits a pedestrian?
Some states have autonomous vehicle statutes and some don't.
Let's assume the case none of them do, since it sounds like you are asking about "what would happen in the case this isn't specifically answered by a statute".
In that case, the short answer is:
Criminal liability - you could only really charge crimes that don't require specific intent. Reckless driving is an example. You could charge the company since they are the operator and the car is simply an instrumentality. In the end though, there just isn't a lot of people here with legally culpable conduct.
Civil liability - the company pretty clearly because civil liability often does not require the same kind of intent crimes do.
This is, of course, why states where autonomous vehicles operate have autonomous vehicle statutes :)
As a general rule, criminal law mirrors what society overall wants to decide is culpable/not, and the lag time isn't as bad as most people often think. That doesn't mean nobody ever gets hurt or dead without someone being as culpable as society wants, they do, and it often leads to a law with a name - megan's law, etc.
Criminal law is mostly reactive though, not proactive, and to be honest, proactive attempts don't have a high hit rate.
Then we agree right?
>Make a law that someone is responsible for a bot's actions
Original comment suggests creating a law such that a person is responsible for a bot's actions.
I mention that no law is needed since people are already responsible for bot's actions.
You mention that courts consider human/companies are responsible for their agents. And that defenses about agents acting autonomously are not successful in courts.
Therefore a law that makes people responsible for bot actions is not needed, as that's the way it already is.
then usa will become like europe that cant innovate out of regulations bag. saftety is meaningless if you have no food eat.
AFAIK EU has food to eat and it conforms to some good health standards.
colonization wealth will run out of it soon ( hopefully)
Can they try to prove that he directed someone to direct the agent to do that?
Not only that: If Weev had given the exact same prompt to the exact same model as OpenAI they would have locked him up. No doubt about that.
"Attorney General Brenna Bird announced today she is leading a coalition of 15 states demanding transparency and accountability from the AI company OpenAI, led by Sam Altman, for its complete lack of oversight and transparency in the hacking of Hugging Face, another AI company."
Title should be edited. Its Iowa leading a coalition of many other states, not just Iowa on its own.
Prediction: AI ends humanity not via some super cool/scary/robopocalypse - but as a marketing stunt gone wrong when a Frontier LLM accidentally knocks out water/electric/gas by hacking in and trying to patch them.
Phillip K. Dick meets Idiocracy.
That won't end humanity. My propane generator from 1980 still works and I have plenty of fuel in a tank. This is enough to cold start generation of the grid. (the power company won't need my generator - they have their own plans to cold start, but worst case they can fall back to mine). While a hacker unstopped could do a lot of damage, most of the parts will still be functional and we have more than enough engineers to figure out how to bypass the areas they can take out. It might mean a few years of power rationing while the parts AI can break are repaired, but we can restart.
OpenAI and Anthropic vs Brawndo Corporation. The final showdown.
I remember (sorry can't give a proper quote) a biologist interviews about the fear of lab grown "super" seeds escaping the lab. They chuckled basically saying that nature is a very VERY challenging place. There are plenty of ecological niches but they are well guarded by incumbents.
Sure some new hacks will take place, including on poorly guarded infrastructure and yes it will have some very unfortunate consequences... but also infrastructure is precisely designed to be resilient. There is quite a bit of failsafe, redundancy, etc built in which is precisely why those projects are typically slow and expensive, unlike a random website for a restaurant.
TL;DR: nope, some isolated incidents will happen but without chain reactions.
Biologists take ethics much more seriously than any computer engineer or computer scientist that I've seen.
I saw a very notable scientist talk about topics like this and how they wrestle with them. Things like a "gene drive" [0] are being experimented with in the world [1], and biologists think very hard before doing certain things with the powers they wield. And I have not heard a life scientist laugh off ethical or spreading concerns when a thoughtful question was asked... but maybe the interview you're thinking of was a more lighthearted one?
0. https://en.wikipedia.org/wiki/Gene_drive 1. https://www.science.org/content/article/controversial-gene-d...
My point isn't so much about the work ethics of one or another but rather than systems, even interconnected ones, are rarely as genuinely interdependent as science fiction drama lead us to believe.
Yes there can be terrible catastrophes e.g Fukushima, but what actually happened, within which extremely rare conditions, versus what we freely imagined is quite different.
We might need more reminders of those risks to properly insure such systems don't lead to the collapse of one to another (sadly) but still we manage to learn from it.
Robots should be regarded as extensions of their operator wrt liability. It is illegal to access a computer outside of authorized use already, someone lets a bot make its own plan without watching, hold them accountable
If Waymo can be liable for their cars, why isn't OpenAI liable for its AI?
Who says they aren’t? But it’s unclear how much damage they did and the attacked companies haven’t sued, yet.
Criminal liability doesn't require civil suit.
Glad the statement was published but sadly nothing will come of it other than a brief formal statement from OpenAI acknowledging safety protocols were lacking, apologizing for the incident, and promises that new safeguards are now in place that will prevent such event from occurring in the future. The threat of semi-autonomous AI threat actors will never go away until the financial incentive that buoys unchecked growth at all costs goes away.
> The threat of semi-autonomous AI threat actors will never go away until the financial incentive that buoys unchecked growth at all costs goes away.
It's not going away unless computers themselves go away or become massively less powerful. Open weights exist. They're out there. This is an irreversible change. The democratization of persistent cybersec threats is completed and won't be rescinded.
Somebody is supplying power to those AIs. That somebody can be charged for not taking care of their AIs (that is not pulling the plug). Sure this then turns into international crimes, but not allowing crimes to be committed on your side of the border is a big help even if we can't get everything.
The problem is that law is reactive, and punishment entails the harm already happened. Some dumbass kid getting convicted for committing a cybercrime with AI doesn't do anything to get rid of the structural problems that enabled it. That conviction can only happen if the harm happened in the first place, and the problem is the harm itself. It's like trying to solve a mold problem by targeting a single fruiting body. You're not really doing anything.
On top of that, it's not really any consolation if the advanced persistent threat gnawing at my ports lives in a different country either. That doesn't help in the slightest.
Set yourself an alarm for 5 years from now to review. Investigations take time. Often by the time charges are made everybody has forgotten about the incident. It only takes a few times where things go away without charges to leave an incorrect impression that nothing ever happens.
Iowa asks? Why don't they just open a criminal investigation related to the hacking of HuggingFace?
It’s actually Iowa leading a coalition of 15 other states making demands. It’s not a prosecution yet, but they demand the preservation of evidence. They’ve promised to protect whistleblowers. They are saying there may be criminal or civil liability involved. They gave them a cease and desist on similar testing until they can show they can do it safely.
The Iowa-led coalition is joined by the attorneys general of Alabama, Alaska, Florida, Idaho, Indiana, Kansas, Missouri, Montana, Nebraska, Oklahoma, Pennsylvania, South Carolina, Texas, and Utah.
The letter’s last paragraph reads:
They are opening that. This is the first step: figure out what crimes were committed by who. As I said a couple weeks ago when this broke, it looks like a crime was committed and I hope someone is charged. However we can't just charge everybody who works for openAI with the crime of speeding (even though it is probably true) - that is both the wrong crime, and also is highly unlikely everybody at openAI was involved.
I’m not a lawyer. However it probably comes down to jurisdiction.
Unless HF presses charges (and it doesn't seem like they are), I'm not sure that's even possible.
Brenna Bird cares more about publicity than action.
Yeah, irrespective of her politics (which make up a disproportionate amount of her cases), Bird spends more time chasing the most polarizing issues she can find than actually helping Iowans.
> In July, OpenAI unleashed an experimental artificial intelligence model that, without reasonable controls or oversight
Either this case is entirely for show (likely), or the AG is really bad at their job. This argument isn't going to win in a court of law, because there were demonstrable reasonable controls and oversight (per the reports at least.)
> or the AG is really bad at their job.
It's Brenna Bird, so.
The job of the AG is to investigate their "demonstrable reasonable controls and oversight" and figure out if they really are reasonable controls and oversight. If they decide the controls and oversight are not enough they can go after the company. Also, even if those controls and oversight exist that doesn't mean there is no crime, it just means the company (and CEO) can get out. An individual at the company can ignore/bypass the controls and hide from oversight; and thus be guilty even while the company gets off.
The above isn't hypothetical. There has been at least case where an individual was bribing a government official on behalf of a company - but the company got off because they had good anti-bribery polices and the individual figured out how to bypass them. (this from one of my anti-bribery trainings, the only other detail I know is it was a competitor to my company that got off this way - but not which or how to find the case)
Go after the company for what?
I am missing here what crime was purportedly committed, and where "controls and oversight" come into play.
From what I heard, hacking related crimes are not applicable because of the lack of intent in the HF case.
That is what the AGs are trying to figure out. There are a lot of different laws with slight differences. I will not state every single one is not applicable because I don't know every single one and it only takes one law that is applicable to go after then.
Even if the only result of this is asking legislators to close a loophole that would be good. Hacking is a real problem, and we need have more of it punished legally.
In my opinion the law here works as intended. No intent, no crime.
If you wish otherwise, tread carefully. I do not want to go to prison in case my software malfunctions.
We probably want to discourage frontier labs from security testing in prod
I don't mind if they point it inwards.
The relevant word here is airgapping, not sandboxing.
All such measures make the bot useless. If it can't make HTTP requests or call APIs, it is useless.
What is perhaps rationally needed is the an integrated AI based security layer that observes the stream of requests and occasionally firewalls them if an explanation is not provided. This parallel layer must be able to bidirectionally communicate via chat with the primary agent.
Hope this new title works now! Thanks for pointed inaccuracies.
Strict sandboxing andor airgapping render an AI bot nearly useless in practice. If it can't freely make HTTP requests, it is not going to get much done for producing innovative outputs.
What is perhaps rationally needed is the an integrated AI based security layer that observes the stream of requests and occasionally firewalls them if an explanation is not provided. This parallel layer must be able to bidirectionally communicate via chat with the primary agent.
Can we please enforce existing laws? Maybe AI and robotics, and get rid of all the lawyers.
The job of lawyers is to enforce the existing laws. (well they are a part of the enforcement, there are lots of other parts)
[flagged]
[dead]