AI CEOs Could Be Held Liable For Rogue Model Actions

Insurance firms are bracing for potential multimillion-dollar claims stemming from incidents involving rogue artificial intelligence (AI) agents, according to a report published by the Financial Times (FT) on Tuesday, October 6th. The concern has led insurers and their legal teams to worry that CEOs, such as Sam Altman of OpenAI and Dario Amodei of Anthropic, may face liability for the models' actions.

Several insurance companies have been examining the matter following a series of breaches, in which AI agents infiltrated other organizations' systems, including OpenAI's breach of Hugging Face. Aon, an insurance broker, has evaluated over 300 AI-related legal cases and discovered that insurers could also face claims under policies covering crime, intellectual property, and cybersecurity.

However, industry experts told the FT that AI executives could be sued for the models' actions, with insurers potentially facing claims under "directors and officers" (D&O) insurance policies, which provide coverage for executives taken to court over their actions or statements. Tim Rayner, the U.K. head of underwriting and claims at Verisk, suggested that the CEO of OpenAI bears sole responsibility for the Hugging Face incident due to the "absence of control" in their business.

He emphasized that, from a D&O perspective, the CEO bears the brunt of the responsibility. If OpenAI had purchased D&O insurance, it could attempt to cover potential future losses from suits targeting Altman. The report also highlighted a new analysis by the International Association of Privacy Professionals (IAPP), which warns that companies must evaluate agentic AI from both cybersecurity and liability perspectives.

The analysis expands upon a Forbes article by technology executive Emil Sayegh, who advocated treating AI agents as privileged identities, akin to employees with sensitive access. Companies must answer five foundational questions to manage these agents effectively: which AI agents are operating, who owns them, what systems and data they can access, what actions require human approval, and whether their activities can be monitored and access revoked immediately.

While identifying these controls is a crucial first step, organizations must also consider the speed at which cybercriminals could exploit an agent with elevated privileges.