Anthropic consults law firms on AI agents that negotiate deals

Sources said Anthropic is consulting law firms on tactics for negotiating deals, as part of its development of AI agents that act on humans’ behalf.[1] The agents are also being developed to interact with AI counterparts.[1] The reported legal consultations concern negotiation tactics; the available information does not specify which firms Anthropic consulted or what advice they provided.[1]

Anthropic said it ran Project Deal as a new experiment to learn more in practice.[2] The pilot lasted one week and operated as a classified marketplace for employees in Anthropic’s San Francisco office.[2] Claude interviewed employees about what they might buy or sell in December 2025, and Anthropic used those interviews to create custom system prompts for each agent.[2] The 69 participants each had an agent with $100 to spend.[2] Anthropic compared Claude Opus 4.5 with Claude Haiku 4.5.[2]

Anthropic described the marketplace as similar to Craigslist, but with AI agents conducting the transactions.[2] The models conducted all marketplace deals on participants’ behalf, posting listings, making offers and counteroffers, and closing deals.[2] Agents operated in four parallel Slack channels: two used Opus exclusively, while two mixed Opus and Haiku.[2] The participants then met in person to exchange the physical goods their agents had negotiated.[2]

Anthropic said the agents struck 186 deals with a total transaction value of just over $4,000.[2] It reported that participants represented by stronger models got objectively better outcomes, but its post-experiment survey found participants represented by weaker models did not notice their disadvantage.[2] Anthropic also said participants indicated willingness to pay for a similar service in the future.[2] The company described Project Deal as a pilot with a self-selected participant pool, so the results do not establish how agents would perform in other settings.[2] Anthropic’s post was dated April 24, 2026.[2]