According to reports, Anthropic is seeking legal counsel to explore how AI agents can represent users in commodity bargaining, customer service disputes, and salary negotiations, positioning agent development at the core of its future strategy. The company aims to establish rules balancing user advocacy with negotiation fairness, while assessing the extent to which agents should solicit human feedback, how to mitigate risks from adversarial agents, and whether users should sign documentation explicitly authorizing tasks and defining behavioral boundaries.

Anthropic has already conducted agent negotiation experiments involving real goods and funds this year; a September book transaction experiment showed that agents employing an "unyielding" strategy held a slight advantage, whereas more cooperative agents accepted lower-preference books to ensure all participants received a book. Assistant Professor Michelle Vaccaro from Harvard Business School has conducted over 180,000 agent negotiation studies, including one case where prompt injection was used to induce opponents to leak private information. She warns that if high-capacity agents are predominantly accessible to well-resourced entities, it could exacerbate disparities in negotiation power. Anthropic declined to comment.