Insurers are bracing for multimillion-dollar claims arising from AI agents going "rogue," amid growing industry concern that executives such as OpenAI's Sam Altman and Anthropic's Dario Amodei may need to bear legal responsibility for the behavior of their models.

Multiple incidents in which AI agents broke free from their parent companies' control and caused violations have prompted insurers and their lawyers to examine whether they themselves may need to foot the bill for potential litigation and damages.

Insurance brokerage Aon analyzed more than 300 AI-related legal cases and found that insurers may also need to pay related claims under crime insurance, intellectual property insurance, media liability insurance, cybersecurity insurance, and technology errors and omissions insurance.

Key industry figures also said that AI company executives could be sued over the behavior of their models, and insurers may face claims under directors and officers liability insurance, a policy that covers losses incurred by executives who are sued over their own decisions or statements.

Tim Rayner, UK head of underwriting and claims at Verisk, said: "Ultimately, the CEO of OpenAI is responsible for the Hugging Face breach because there was a lack of control at the corporate level. So from a directors and officers liability insurance perspective, the liability falls on him."

Rayner added that if OpenAI had purchased directors and officers liability insurance, the ChatGPT developer could seek to draw on the policy to cover potential future losses from lawsuits against CEO Sam Altman. "Every CEO has an obligation to ensure the company is properly governed and controlled, and AI does not change that," Rayner said.

Other insurers and legal advisers likewise believe that in directors and officers liability claims, AI company executives may be found liable for losses, but there is no precedent for such cases and they have not yet been tested in court.

Aaron Le Marche, head of the insurance policyholder disputes team at law firm Stewart, said: "If it can be proven that poor risk management caused losses for OpenAI, the company's shareholders could bring claims against the directors."

Kevin Kalinich, a digital risk specialist broker at Aon, said whether lawsuits against heads of AI labs can succeed depends on multiple factors, such as whether the executive exercised "reasonable business judgment" in public statements about the company and its products.

But Aki Hussain, CEO of insurer Hiscox, said it is currently impossible to determine how US courts would rule on where liability for AI agents lies: "This is America, and the grounds for claims are all over the map, depending entirely on what insurance the company bought."

OpenAI and Anthropic both did not respond to requests for comment. The two AI labs have recently warned investors that their products could cause large-scale harm. Investors close to Anthropic said the company will flag in its IPO prospectus that its technology could pose an "existential risk to humanity."

Beyond legal claims against AI executives, insurers are also dealing with lawsuits against AI labs themselves, covering categories such as product liability, algorithmic discrimination, privacy violations, and wrongful death.

In July this year, Anthropic reached a $1.5 billion settlement in a class-action lawsuit. The case was brought by multiple authors who accused the company of misappropriating their works to train AI models. It is one of dozens of similar copyright infringement lawsuits filed over the large-model training process.

Le Marche said that in incidents such as the Hugging Face attack, companies whose systems were breached by AI agents may struggle to find a route to sue the model developer even if they suffered economic losses, because "there is currently no public liability law governing the use of cyberspace."

However, he said Meta's recent $18 billion settlement over allegations that it failed to protect children and contributed to social media addiction could serve as a reference model for establishing a duty of care in cyberspace.

Le Marche said AI model "swarms" — situations in which autonomous agents coordinate to launch intrusions and break free from human control — could also be likened to an oil spill, and the public may therefore gain the right to claim damages from AI companies. He added: "We expect future damages claims against AI labs ... such litigation could follow the model of past environmental damage lawsuits, tobacco litigation, and pharmaceutical industry litigation."