Anthropic consults law firms to set fair negotiation standards for AI agents bargaining on users' behalf

According to the Financial Times, Anthropic is consulting law firms to develop negotiation standards for AI agents covering haggling, complaint handling, and securing better contract terms. Anthropic's own experiments show stronger models negotiate better deals, raising concerns about widening social inequality. The company is also recruiting Conceptual Reasoning Fellows with legal backgrounds for a four-month program testing Claude on judgment-dependent questions in law, philosophy, cognitive science, economics, and political science, at $100K–$167K. The root issue is a data gap. Math and code have abundant, verifiable data, making improvement relatively easy; the humanities lack such data partly because correct answers are often ambiguous. Anthropic technical staffer Justin Curl notes that law is "notoriously hard to define quality" for, so the company needs experts to build test cases and evaluations. For teams building agent products, fairness standards for negotiation agents will directly determine whether bargaining, complaints, and contract talks can be automated; for teams relying on Claude for legal or contract judgments, the completeness of evaluation standards and test cases decides whether outputs can be trusted in real business.