Who is legally liable when an AI agent goes rogue?
When autonomous AI agents behave unpredictably and cause harm, determining legal liability is complex. Currently, there is no specific federal AI agent liability law, so existing legal frameworks are applied. The AI agent itself cannot be held liable, as it is not a legal entity. Liability typically falls on the "developer" (the maker of the AI) or the "deployer" (the user), depending on the facts and circumstances. A negligence analysis under standard tort law may apply. For instance, if a deployer gives a reckless instruction, such as demanding quick money without safety parameters, they could face significant liability, including potential criminal charges under statutes like the Computer Fraud and Abuse Act.
The situation is complicated by open-source models, where licenses often disclaim liability, and by the unclear division of responsibility between developers and deployers. An analogy is drawn to self-driving car accidents, where both the manufacturer and the human operator can share fault. In cases of severe harm, such as using AI to create bioweapons, liability for developers depends on jurisdiction; the EU's AI Act imposes responsibilities, while U.S. law offers less clear grounds, similar to platforms being shielded for user-generated content under Section 230.
Even if Artificial General Intelligence (AGI) is achieved, the expert argues against making AGI itself a legally liable entity, as it lacks personhood, assets, or a meaningful way to provide remedy for harm. Instead, accountability should rest with the responsible human or corporate entities behind the technology.
cointelegraphHá 1h
