Anthropic asks law firms how AI agents should negotiate

Picture your agent pushing for a better contract, chasing a complaint or haggling over a price while the other side’s agent wants the opposite, The Decoder explains. How suspicious should an agent be of the one across the table, short of turning “paranoid”? And how often should it check back with its human? Anthropic is weighing both, one source told the FT. The company declined to comment to the paper.

There is a fairness worry underneath. Anthropic’s own experiments found that stronger models strike better deals, The Decoder notes, so people with better agents could pull ahead. It is also hiring lawyers as Conceptual Reasoning Fellows, for four months, to put questions to Claude where the answer turns on judgement. The pay runs from $100,000 to $167,000.

For sellers, it points to a market where a customer’s agent bargains with yours over price and refunds.