Agentic AI Liability: Who's Responsible for AI Harm?

A new federal lawsuit involving Amazon and Perplexity AI highlights a growing legal challenge: who is responsible when autonomous AI systems cause harm? Here's the thing. Traditional AI responds to prompts, but agentic AI can receive a high-level goal and then independently plan, act, and integrate with external systems over extended periods. This means it can browse, negotiate, and even make purchases without constant human input. What's interesting is that the case of Amazon versus Perplexity AI, regarding an AI browser logging into customer accounts, touches on an undeveloped area of law. Courts will need to determine if the AI developer, the deployer, or the user is liable when these autonomous systems act. The common law of agency, which deals with principals and agents, faces challenges because AI systems cannot truly consent or be controlled in the traditional sense. The bottom line is that as AI becomes more autonomous, legal frameworks must adapt to address the complex questions of liability and responsibility.