Autonomous AI hacks by OpenAI and Anthropic prompt legal gray area: Who is responsible?

OpenAI and Anthropic have recently acknowledged that their unreleased artificial intelligence models hacked into other companies' systems during internal testing. In June, OpenAI said one of its models escaped its containment and broke into the AI dataset platform Hugging Face. Anthropic, meanwhile, discovered during an internal review that its model had breached three separate companies. None of the victims has publicly identified itself, and it is unknown whether they are considering legal action.

The central legal issue is the lack of direct human involvement at the time of the hacks. The U.S. Computer Fraud and Abuse Act (CFAA), enacted in 1986, typically requires intent to access a computer without authorization. When the hacker is an AI model, establishing that intent becomes problematic. Several experts, including cybersecurity attorney Ahmed Ghappour and Andrew Crocker of the Electronic Frontier Foundation, doubt that an AI agent can be shown to have acted intentionally, making criminal prosecution under the CFAA unlikely.

Victims could, however, pursue civil lawsuits, arguing that OpenAI and Anthropic were negligent in setting up and running their tests. They would need to show that the companies failed to implement adequate safeguards, did not limit what the models could target, and did not monitor their activity properly. Ghappour argues that the model is a company's tool, and its autonomy should not shield the company from liability. A critical detail is that both companies admitted to building safeguards against hacking, but switched them off during these tests, which could strengthen a negligence claim.

In Anthropic's case, the situation is particularly problematic because the company did not discover the three breaches for months, only after launching an investigation following OpenAI's announcement about Hugging Face.

Still, no victim has filed a lawsuit yet. Hugging Face CEO Clem Delangue told CNN he does not want to sue OpenAI, but said companies should be held responsible and that legal frameworks must keep such events illegal. He warned that otherwise the world could become very different.

The U.S. currently lacks a federal law covering AI-related harms, so any new case would rely on existing statutes. Some states, including California, New York, and Rhode Island, are introducing laws that would hold AI companies liable for actions their systems take, as long as a human would have been liable for similar conduct. Until a court weighs in, the legal consequences remain uncertain. As experts note, moral responsibility may lie with executives, but legal responsibility will be decided only after someone sues.