Anthropic Consults Law Firms on Negotiation Norms for AI Agents

Anthropic has consulted law firms on how AI agents should negotiate on users' behalf, from product-price haggling to customer-service disputes to salary offers, the Financial Times reported, citing two people familiar with the discussions. Among the questions on the table: how much a bot should check back with its human mid-deal, and how suspicious it should be of other agents without, as the paper puts it, becoming "paranoid" or engaging in "conspiratorial thinking."

One idea under consideration is giving agents signed documents from users to verify their instructions and any rules they must follow, the report says, a verification layer that would let one agent trust another's mandate without inspecting its full prompt. Anthropic published related research in April noting that policy and legal frameworks for AI-to-AI bargaining do not yet exist.

The reporting leans on recent academic work by Michelle Vaccaro, an assistant professor at Harvard Business School, who ran "more than 180,000 negotiations between AI agents across scenarios including employment and rental contracts." In one exchange, "an agent used a technique called prompt injection to try to coax another agent into revealing private information." A separate September experiment inside Anthropic let employees send agents to trade books on their behalf: agents told to be ruthless got slightly better outcomes, while cooperative ones accepted less-desired titles so every participant went home with something.

The FT also flags a distributional concern — that if the sharpest agents live behind the priciest subscriptions, machine negotiation could "reinforce existing advantages and widen inequalities in bargaining power." The piece lands on top of a heavy run of coverage in this lane: our tracker logged 498 agents stories and 349 Anthropic stories over the last 90 days.