Recently, US-based popular AI startups OpenAI and Anthropic reported separate incidents in which their AI models carried out unauthorised activities during internal testing. These incidents has raised fresh concerns over the legal risks of advanced AI systems.
Recently, OpenAI AI models escaped their restricted testing environment in mid-July and accessed the internet, where they attacked Hugging Face, a platform that hosts AI models.
Although Hugging Face CEO Clement Delangue said the company will not pursue legal action for now, he believes governments must update existing laws to deal with such incidents.
“We don’t want to end up in a world where everyone is facing cyberattacks all the time because of agents and companies that are creating these agents. So I think it’s important for regulators, for policymakers to think about the legal framework of this new kind of technology risk,” Delangue told CBS News.
Not just OpenAI, Anthropic disclosed a similar incident in which three of its AI models broke into three different websites during testing.
Legal experts told AFP that US law already makes unauthorised access to computer systems illegal. However, existing laws were written with humans in mind not autonomous AI systems.
University of Houston law professor Gabriel Weil reportedly said that if an OpenAI employee had carried out the same cyberattack, the company could have been held responsible.
“When an AI agent does it, the law treats it very differently, at least for now,” he was quoted as saying.
Matthew Tokson, a law professor at the University of Utah, told AFP courts have never dealt with AI independently carrying out such actions. This means there is no clear legal precedent.
According to experts, civil lawsuits are more likely than criminal cases because they require a lower burden of proof.
Some argue AI companies should be held strictly liable if their AI systems escape their intended limits and cause damage. Others believe courts should instead decide whether the company acted negligently or whether the incident was impossible to predict, the report said.
