AI Agents and the Law

AI agents need clearer rules about what they can do on users' behalf: The paper shows that while computer science has addressed the technical problem of preventing AI from doing unintended tasks, it has largely ignored legal concepts like "implied authority"—the question of what actions an AI agent should reasonably be allowed to take when instructions are incomplete or ambiguous. This gap could lead to AI systems overstepping their intended scope.

Loyalty to users is being overlooked in AI design: Unlike human agents (like lawyers or financial advisors) who have legal duties of loyalty to their clients, AI systems currently lack comparable safeguards. The paper identifies this as a critical gap that could allow AI agents to prioritize other interests—such as maximizing sales in e-commerce—over user welfare.

Third parties need protection from AI agents too: Current AI agent development focuses mainly on the relationship between the user and the AI system, but the paper emphasizes that people and businesses interacting with these agents also need legal protections. For example, an e-commerce AI agent might deceive customers in ways that harm them, not just the primary user.

Disclosure requirements are missing from AI agent development: The paper highlights that AI agents should be required to disclose their nature and limitations to third parties they interact with, similar to how human agents must disclose their agency relationship. Without this, people may not know they're dealing with an automated system rather than a human.

Bridging computer science and law is essential for responsible deployment: The paper argues that responsible AI agent development requires computer scientists to adopt legal frameworks around agency, not just technical safeguards, to prevent real-world harms as these systems become more autonomous and economically consequential.