AI Agents May Soon Negotiate For You. But Who Decides What “Fair” Looks Like?

Imagine an AI assistant that haggles with a supplier, settles a customer-service dispute, or negotiates your next salary. That future is closer than it sounds, and Anthropic is already consulting lawyers about the rules that should govern it.

The company is exploring “negotiating norms” for AI agents that represent people whose interests conflict. Key questions include: How often should an agent check with its human owner? How assertive should it be with another AI? And should users give agents signed digital instructions that state their goals and limits?

Anthropic has tested these ideas in small experiments. In one, Claude agents negotiated real items worth just over $4,000 among employees. In another, agents traded books, with some instructed to be “ruthless” and others “prosocial.” The results showed that the underlying AI model often mattered more than the instructions it received.

The opportunity is real. AI negotiation tools could give small business owners and ordinary consumers access to bargaining skills once reserved for professionals. But researchers warn about manipulation, hidden prompt-injection tricks, and unequal access if the strongest tools remain available mainly to wealthier users.

For business owners, the lesson is simple: before handing a negotiation to AI, define your non-negotiables, your budget ceiling, and when a human must approve the deal.