Anthropic is seeking legal advice on how AI agents should haggle, settle disputes, and bargain on behalf of the people they represent, according to a Financial Times report.
The company has approached law firms to understand how lawyers conduct negotiations, balancing their clients’ interests against the expectation of fair dealing. The question is becoming urgent as personal agents grow more capable and start encountering one another, with goals that may conflict.
Anthropic is weighing how often an agent should check back with its user mid-negotiation, and how suspicious it should be of another AI without tipping into paranoia. One idea under consideration involves giving agents signed documents from the user confirming their brief and any rules to follow, the FT report explains.
The work sits at the centre of Anthropic’s strategy as it prepares for an initial public offering that investors believe could be the largest listing on record. Competitors are moving in the same direction: Meta’s Muse and OpenAI’s Dots are already in the market.
Anthropic’s own experiments have offered early signals, further states FT. In September, employees used agents on a virtual trading floor to buy and sell books, ranking their preferences beforehand. Agents instructed to be ruthless finished slightly ahead, while cooperative ones made deliberate sacrifices so that every participant ended up with a book.
Anthropic said in April that the legal and policy frameworks for AI systems that transact on our behalf “simply don’t exist yet”, adding that society would need to move quickly. The company declined to comment to FT on the matter.
