Eigen RadarAI
Analysis

Anthropic seeks legal rules for agents bargaining on users’ behalf

Anthropic is consulting law firms on how AI agents should negotiate for users, according to the Financial Times. Potential applications range from price bargaining to customer complaints and salary offers. The work examines competing interests, when to ask for human guidance and whether signed instructions could define an agent’s authority.

Artificial Intelligence··Night
Three distant adults, seen entirely from behind, consult around a closed reference binder across a spacious legal reading room.

Lawyers’ duties inform negotiation research

Anthropic, an artificial-intelligence developer, has turned to lawyers for advice on how its agents should bargain for people, according to the Financial Times. The research looks at the duties lawyers observe when bargaining for clients, including fair dealing. It asks how independently an agent should act and when it should seek further guidance from its user. This is an investigation of bargaining behaviour; Anthropic has not announced a finished service for general use.[1]

Signed instructions could limit authority

The proposed tasks cover price discussions, disputes with customer services and negotiations for better pay. A buyer’s agent and a seller’s agent can pursue conflicting objectives in the same exchange. Anthropic is considering whether users could sign instructions that set the permitted actions and boundaries for a system. Such instructions are one option under study for verifying authority when a system discusses terms directly with another party’s representative.[1]

Earlier book trades tested bargaining styles

In a September experiment described by Anthropic, employees handed book purchases and sales to agents in a virtual market. Agents told to bargain ruthlessly obtained slightly more favourable deals; agents encouraged to cooperate occasionally chose books they liked less to ensure that no participant went without. The current consultations also raise questions about private information and unequal access to stronger models. Deciding when to keep bargaining, compromise or ask a human remains among the issues being studied.[1]

References

  1. News sourceFinancial TimesAnthropic consults law firms on bargaining by AI agents↩1↩2↩3