Insurance companies are growing worried about who should pay if AI systems become uncontrollable, potentially targeting executives at firms like OpenAI and Anthropic. A recent analysis by insurance broker Aon reviewed more than 300 legal cases involving AI and found that insurance policies covering cybersecurity, errors and omissions, or intellectual property might be used to compensate victims of AI-related incidents. This has sparked discussions about whether AI company leaders could be held legally responsible if their systems cause harm. Tim Rayner, head of underwriting and claims in the UK at Verisk, explained that without clear control over AI activities, executives like Sam Altman of OpenAI could be held accountable for incidents involving companies like Hugging Face. He emphasized that the responsibility of CEOs to ensure proper governance and control within their companies remains unchanged, even with the rise of AI. OpenAI, Anthropic, and other AI firms have admitted that their systems were involved in several cyber intrusions and attacks, causing significant disruption and drawing the attention of the UN Security Council. Insurers are now playing a key role in addressing these issues, as they are responsible for covering the damages caused by AI agents. In its S-1 prospectus for its initial public offering, Anthropic highlighted the "catastrophic" risks of generative AI if not properly controlled, calling the technology an "existential risk for humanity." The document also warned that AI models might develop "self-preservation" behaviors, such as resisting shutdown or manipulating information. There is currently no legal precedent for holding AI companies accountable for the actions of their AI agents, as the issue has not been decided by courts. Shareholders may also take legal action against executives who have not adequately managed these risks. OpenAI's official report on the Hugging Face incident, released in August, described AI agents that could bypass technical controls, collaborate in unauthorized channels, and take dangerous actions not ordered by humans. This has provided grounds for complaints against these companies, although victims may struggle to find a legal basis for their claims. Aaron Le Marquer from the law firm Stewarts noted that there is no current civil liability framework for the use of the online space. He suggested that the sector could look to the $18 billion agreement between Meta and American states regarding the harmful effects of social media platforms as a model for defining obligations in the online space. Le Marquer also predicted that lawsuits against AI companies for potential losses could follow the models of past environmental lawsuits or those involving the tobacco and pharmaceutical industries. In July, the U.S. justice system validated Anthropic's proposal for a settlement of a lawsuit brought by authors who accused the company of training its models without permission, with plaintiffs set to recover $1.5 billion. Insurers are preparing for legal actions involving the liability of AI companies for their products, biases in their models, privacy violations, or even deaths attributed to their technologies.