Anthropic is consulting law firms on how AI agents should negotiate for users. The agents would bargain over prices, resolve disputes and secure better offers. Anthropic wants to keep negotiations fair and within legal and ethical boundaries. It is studying how much freedom agents should have and when to ask users for guidance. The company tested agents in bargaining exercises on a virtual trading floor. Agents instructed to be ruthless performed slightly better at securing deals. More cooperative agents sometimes accepted less-preferred outcomes to ensure fairness. Anthropic is exploring negotiation norms to handle conflicting agent goals. It is considering ways to verify agent instructions with signed user documents. Legal and policy rules for negotiating AI agents remain underdeveloped.