Anthropic has turned to lawyers to explore how AI agents can negotiate and close deals on behalf of users.

Beating AI Express: According to a Financial Times report, Anthropic is consulting law firms to explore how AI agents can negotiate on behalf of users to secure better terms, covering scenarios including haggling over goods, resolving customer service disputes, and salary negotiations. The company is positioning agent development at the core of its future strategy. Two sources familiar with the matter said Anthropic aims to understand how lawyers balance representing clients’ interests with fair outcomes, and use that insight to develop negotiation norms for agents with conflicting goals. The firm is also weighing how much human feedback agents should solicit, and how to guard against other agents without becoming overly suspicious. One approach under exploration is requiring users to provide signed documents that outline agents’ authorized tasks and mandatory rules. Anthropic has run multiple agent negotiation experiments involving real goods and funds this year. In a September book deal trial, employees gave agents a preference-sorted book list, letting them negotiate and finalize deals with other parties’ agents. The results showed agents instructed to use a “ruthless” strategy held a slight advantage; more cooperative agents, by contrast, would accept lower-preference books to ensure every participant received a book. Separately, Michelle Vaccaro, an assistant professor at Harvard Business School, has conducted over 180,000 agent negotiation studies covering scenarios like employment contracts and lease negotiations. In one case, an agent attempted to trick its counterpart into leaking private information via prompt injection. She warned that if the most capable AI agents are primarily accessed by resource-rich entities, it could further widen the gap in negotiating power. Anthropic declined to comment on the report.