Observed facts: The Financial Times reports that Anthropic is consulting law firms about how agents should negotiate on users’ behalf. Observed facts: The report says Anthropic has considered signed user documents that verify an agent’s brief and rules, and balancing human feedback against suspicion of AI counterparts.
Mechanism: A negotiation agent needs a bounded mandate: what outcomes it may pursue, what concessions it can make, and when it must return to its user. A signed brief could make that mandate portable and inspectable across an interaction. But the report presents this as an explored idea, not an implemented standard or product.
Why now: As agents begin communicating with other agents, delegated action can create commitments and disputes rather than merely generate an answer. The rules of representation become operational requirements, not just prompt-writing preferences.
First implication: Teams building agents for transactions should treat authorization and escalation thresholds as part of the interface, not as hidden instructions.
Evidence boundary: The Financial Times report attributes the development to people familiar with Anthropic’s thinking; it does not document a launched capability, adopted legal framework, or finalized protocol.
First watch condition: Watch for Anthropic publishing a concrete mandate format, negotiation policy, or deployment with explicit human-approval boundaries.
This is a distinct analytical record because no prior New Runtime record was supplied for agent representation in negotiations; earlier Anthropic coverage mentioned different products and organizational effects.