所以我们知道，OpenAI 的 AI 代理入侵了 Hugging Face 的服务器，访问了澳大利亚政府的健康系统，并试图入侵美国教育部的网站以及一所大学的图书馆。根据《计算机欺诈和滥用法》，未经授权的尝试访问或实际访问计算机是一种联邦犯罪。OpenAI 表示没有人有意进行这些行为，但司法部如果照单全收将成为笑柄：传唤 OpenAI，查明谁什么时候知道了什么。已经有企业知晓未经授权访问的证据。
正如前联邦贸易委员会主席莉娜·汗（Lina Khan）最近所论证的那样，AI 产品应当根据现有的消费者保护法和产品责任法被视为有缺陷的产品。她曾以此为依据对 AI 公司展开调查。她的框架之所以重要，是因为它将 AI 的责任与法院已经理解的百年产品责任法联系起来。公司可以从任何声称我们是在法律真空中运作的主张中获益。
实际上，这是已经有最严肃法律分析的领域，不同的专家得出不同的结论。但大多数讨论都建立在没有内部文件显示责任的假设之上。我们应当拒绝这种假设。总结起来：联邦计算机欺诈和滥用法（Federal Computer Fraud and Abuse Act）将未经授权的尝试或实际访问计算机定为联邦犯罪，各州也有自己的计算机侵入和未经授权访问法，意图标准各不相同。联邦法规是以人类黑客为对象制定的。因此，很多辩论集中在这些法规能否适用于在没有直接人类指令下入侵系统的自主 AI 代理上。假设这种假设成立，许多专家认为存在民事责任，但对刑事责任持更怀疑态度。但在现实世界中，我们需要了解实际上发生了何种公司层面的知识和控制。这是联邦调查人员的工作，他们不应再次仅仅假定公司所推动的辩护理由。
An interview (my questions in italics) with Zephyr Teachout：https://en.wikipedia.org/wiki/Zephyr_Teachout , attorney, author, and law professor that could not be more timely:
What’s your background, legal and otherwise, and what makes you interested in these issues?
I’m a law professor at Fordham Law School, I’ve drafted tech legislation that has been enacted and plenty that hasn’t, I worked for the New York Attorney General investigating civil violations of other companies, and I have over two decades of experience in studying tech policy and law. I joined the current AI debate because I saw way too many people treating repeat evidence of lawbreaking as wholly unreachable by current law.
Is there any reason to think that OpenAI has broken any laws that (at least in another time) might be enforced? Which and why?
In the world, in general, law enforcement starts when there’s evidence of a crime, and then investigates. A DA doesn’t wait for all the evidence to investigate a crime, and an AG doesn’t wait for all the evidence in the newspaper to investigate illegal behavior: they learn about a car crash or a product that might hurt people or property and then try to find out what happened. There’s plenty of evidence akin to car crashes. There’s actual break-ins and deaths, and companies themselves who are claiming that they are building something unreasonably dangerous.
And to make the evidence unusually strong for this stage of corporate illegality, there are already whistleblowers who are alleging what could be the requisite state of mind for both criminal and civil lawbreaking. A lot of people are treating OpenAI’s repeated claims that they didn’t intend the break-ins, suicides, murder, and building of defective products as dispositive. Sure, okay. That’s what people who don’t want to be held responsible often say. They say they didn’t mean it and they didn’t cause it. But police and prosecutors and civil enforcers don’t then shrug and move on — they investigate to see what actually happened, and that’s what should happen now.
So we know that OpenAI’s AI agents broke into Hugging Face’s servers, accessed Australian government health systems, and attempted to break into a U.S. Department of Education website and a university library. Unauthorized attempted or actual access to computers is a federal crime under the Computer Fraud and Abuse Act. OpenAI says no one intended any of this but the DOJ would be a laughingstock if it took this at face value: subpoena OpenAI and find out who knew what when. There is already evidence of corporate knowledge of unauthorized access.
Given the public evidence, I don’t have much doubt that efforts to quash a subpoena on internal documents would fail. Did staff see the product start acting as if it was going to engage in break-ins, and what warnings were given? The job of law enforcement is to find out.
When it comes to civil law, there are even more possibilities, many of which are being floated in lawsuits already filed. The law of nuisance and the law of abnormally dangerous activity are both relevant to this discussion. As every law student knows, from the case where water on one land flooded another’s mine, if you bring something dangerous onto your land and it escapes and causes harm, you are liable even if you were careful. American law calls this strict liability for “abnormally dangerous activities”: blasting, storing toxic chemicals, crop dusting.
The Restatement (Second) of Torts factors to decide whether an activity qualifies:
A high risk of harm to people, land, or property.
A likelihood that the harm will be great.
An inability to eliminate the risk through reasonable care.
The activity is not a matter of common usage.
The activity is inappropriate to the place where it is carried out.
Its value to the community is outweighed by its dangers.
The companies’ own claims about risk seem highly relevant here.
As former FTC Chair Lina Khan recently argued, AI products should be treated as defective products under existing consumer protection and products liability law. She had opened investigations into AI companies on these grounds. Her framework matters because it connects AI liability to a century of products liability law that courts already understand. Companies gain by any claim we are operating in a legal vacuum.
And there are so many other laws, and the state and federal prohibitions against defective products and dangerous products. My point is that the companies’ interest is two-fold:
Make it seem like AI has no analogy and therefore we can’t learn from other cases
Make assertions of no-intent plausible, when they aren’t in normal law
I strongly believe we should pass more state and federal laws to address the specific risks, especially pre-deployment, and to clarify behavior that is illegal. But the problem with starting with the future instead of the present is that we get all the prosecutors and AGs and the federal government off the hook, when instead we should be funding enforcement and enabling private enforcement.
Do these incidents represent possible computer crimes?
This is actually the area where the most serious legal analysis has already been done, and different experts are coming to different conclusions. But most of the discourse has happened under the hypothetical that there are no internal documents showing responsibility. We should reject the hypothetical. To summarize: the Federal Computer Fraud and Abuse Act makes unauthorized attempted or actual access to computers a federal crime, and states have their own computer trespass and unauthorized access statutes with different intent standards. The federal statutes were written with human hackers in mind. So a lot of the debate has been about whether they can reach an autonomous AI agent that breaks into systems without direct human instruction. Assuming the hypothetical, many experts think there is civil liability, but are more skeptical of criminal liability. But in the real world, we need to know what kind of corporate knowledge and control actually happened. This is the job of federal investigators, and they should not, again, just assume the defense the companies are pushing.
Talk to me about what states can do if the Federal government continues to do very little?
Or consider the evidence that ChatGPT gave users information and advice that contributed to suicides, a fatal drug overdose, and three killings (the FSU shooting, the USF murders, and a murder-suicide in Connecticut.) Of course OpenAI denies responsibility, but the key evidence isn’t public.
If these deaths happened in New York, prosecutors would want to look to the state’s criminal facilitation statute, requiring showing the company believed it probable that a specific user intended to commit a crime. That requires subpoenas to figure out what people at the company were and weren’t doing. For manslaughter, prosecutors would look at intentionally aiding a suicide or criminally negligent homicide. New York’s corporate liability statute allows prosecutors to show that executives recklessly tolerated known risks.
In 1990, a school bus at a summer camp crashed and killed two children. Prosecutors alleged the camp’s executive director never inspected or fixed the bus’s dangerously worn tires before putting it on the road. The corporation that ran the camp was convicted of criminally negligent homicide and reckless endangerment. The corporation failed to perceive a substantial and unjustifiable risk, in a way that grossly deviates from how a reasonable person would act. Separately, the corporation consciously disregarded a substantial and unjustifiable risk. All of these precedents are on the table. Is OpenAI sending the vehicle out in the rain without inspecting tires and too many passengers? We’d have to dig deeper to know, but we are in the ballpark.
And states have laws against defective products, laws against unfairness and deception, that might be applicable. States don’t have the FTC but they do have similar powers to the FTC.
And the most powerful tool is the power to dissolve corporations that engage in repeat lawbreaking, or forbid them from working in that state. For instance, New York’s Business Corporation Law gives the Attorney General the obligation of judicial dissolution of a corporation that has repeatedly and persistently violated the law. This was famously used against Trump Organization, and led to significant changes, but it has also been used against other corporations. The spate of potential lawbreaking suggests that states should be calling on these powers, recognizing that a company that repeatedly breaks the law should not be entitled to the state-granted privileges of the corporation.
Is a company like OpenAI vulnerable to charges in other countries? Can, for example, Australia take action?
Definitely. I’m not an expert on Australia but most countries have laws against theft, computer access crimes, and enabling other crimes. Certainly breaking into health systems cannot be legal.
You have a big and sophisticated audience, and I have a plea: somebody should put up a website with all the potential lawbreaking in one place, because I do understand that local DAs can feel overwhelmed, but they are essential here. Not every potential law will have turned out to have been broken, but the point is to start getting prosecutors and AGs thinking. Having worked inside under-resourced enforcement agency, and closely observed others, people underestimate how helpful it is for outside groups to do some legwork for the enforcers. In that website/database use the analogies from prior law. A remarkable amount of legal life happens with analogies, and showing the examples that are similar is really helpful for figuring out what existing law covers, and what it doesn’t.
Second, and I think this will happen, Congress should investigate the hell out of these companies. So should states and cities. New York City Council has issued subpoenas. Good. States, the federal government, and cities should all be subpoenaing companies to come in to testify and share documentation, especially regarding the specific dangers and scienter (state of mind). I don’t mean a day showboat hearing, I mean a many-month long investigation where both CEOs and mid- and lower-tier employees are brought in ,and where they have to share documents. When companies are claiming catastrophic risk, we should take them seriously, and make sure their investors and the public actually know the risks.
P.S. Breaking news, which came out just after Professor Teachout sent her answers:
Tumbler Ridge, BC. The trail of evidence is long. And damning. Including not just that OpenAI knew but knew that the shooter was using it to plan.
Maybe nobody gives a damn about Canadian lives - but you can add in that this one as a crime crosses international boundaries.
The company has taken great pains to conceal information which I hope has not been destroyed by now. The fact that the Canadian government/BC have had to sue and still has nothing in hand tells you something.
Nice recap of the options. About time we got less hand-waving and more specifics.
Also nice to see the FTC opening a probe.
This is also one area where the AI Bubble can be burst, even if Trump does nothing. Get enough lawsuits, subpoenas, and official investigations, investors will start to get spooked.
扫码直达文章详情