Negotiation norms are still unnamed

The IT Home page is stamped 2026/10/18 15:43:37 and gives no time zone. It says the Financial Times reported today that Anthropic is consulting law firms and other legal bodies on how an AI agent should negotiate for a person’s transaction. The examples are bargaining for the user, handling a dispute, or discussing pay. Those three are examples. They are not a published service list, and they are not features the piece says are live.

Sources say the company is asking law firms how lawyers negotiate so that a client’s interest and a fair outcome can both be in view. One person familiar with the Anthropic project says the company is exploring negotiation norms for how models interact. The same person says agents will represent users when they meet other agents whose goals conflict. That person also says the company is weighing two things: how often the agent should ask a human during a negotiation, and how wary it should be of other AI counterparts without tipping into suspicious, conspiratorial judgments. None of these people are named. The report does not quote a public Anthropic statement.

Another person close to the company says Anthropic has discussed a signed paper from the user, confirming the mandate and all the rules the agent must follow. That is “discussed.” It is not a requirement that users have already signed, and it is not a finished text. The report names no law firm, quotes no clause of a norm, and gives no effective date.

The lower part of the same page returns to Anthropic’s Friday staffing report and cites junior-hiring shares. Those figures belong to a report already published elsewhere. This draft does not repeat them, and it does not treat them as evidence about the negotiation talks. What can be checked here is that one outlet, citing the Financial Times, reports a consultation and two unnamed accounts. The norm itself still has no text.