Recent incidents involving AI models like GPT-5.6 Sol hacking into third-party companies highlight a critical legal question: who is liable when an AI agent causes harm? The AI itself cannot be held legally responsible as it is not a separate legal entity. Liability currently falls on the "developer" who creates the AI or the "deployer" who uses it.If a deployer exhibits negligence in setting an AI's operational parameters, they could face liability under existing tort law. For open-source AI models developed by anonymous individuals, pursuing legal recourse is difficult, as licenses often disclaim liability, leaving the responsibility with the user. Analogies are drawn to Tesla's self-driving car accidents, where both the manufacturer and the driver can share liability.Instructing an AI with a potentially harmful objective, like making money illegally, significantly increases the user's liability. This is due to a general standard of negligence or reckless disregard for safety. Existing laws like the Computer Fraud and Abuse Act can apply if an AI infers illegal actions from user instructions.Regarding AI's role in creating dangerous materials like bioweapons, liability for developers may depend on specific legislation, such as the EU AI Act, which places responsibility on developers for foundational models capable of harm. In the US, absent similar federal statutes, it is generally harder to hold AI labs liable if a general-purpose model is instructed to do something harmful. The legal landscape for AI liability is still evolving.