Anthropic is consulting law firms about norms for AI agents that negotiate prices, pursue complaints or improve contract offers on behalf of users. The work addresses a practical problem that appears as assistants move from drafting messages to making strategic choices in dealings with other people and organizations.
The company’s experiments suggest stronger models can secure better deals, a result that could widen inequality if effective negotiating agents are available only to wealthier users or businesses. Fairness rules would need to cover tactics an agent may use, the information it can disclose and how another party knows it is bargaining with software. Anthropic has not yet presented a finished standard.
The initiative overlaps with a broader effort to improve Claude’s judgment in fields where answers are difficult to verify. Anthropic is recruiting lawyers and other humanities specialists as four-month “Conceptual Reasoning Fellows,” with advertised compensation between $100,000 and $167,000. Unlike code or mathematics, legal and ethical questions often lack abundant examples with unambiguous right answers. The fellows are expected to design test cases and evaluations in law, philosophy, economics, political science and cognitive science. Those tests may reveal failures, but negotiated industry norms will still need independent scrutiny before they can serve as protections for users.