Anthropic Asks Lawyers to Teach AI Agents to Negotiate

According to the Financial Times, Anthropic is consulting with law firms on how AI agents should conduct negotiations on behalf of users—bargaining over prices, resolving disputes, and securing better salary offers.

Compiled by Sergey KostenchukPublished 2026-10-18Updated 2026-10-18

2026-10-18 Business Anthropic

💼 Anthropic Asks Lawyers to Teach AI Agents to Negotiate

According to the Financial Times, the company is consulting with law firms on how agents should bargain over prices, resolve disputes, and secure salary offers. The company plans to limit agent authority through “signed documents” from the user.

🌍 Anthropic is shaping the norms for agentic deals: who is liable if an agent “binds” a user, and how to verify its mandate. This sets security standards—protection against prompt injection—for everyone building autonomous trading agents.

👤 In September 2026, Anthropic tested Claude agents in a virtual book-buying and selling scenario: “uncompromising” agents got slightly better terms, while cooperative ones got worse. According to Harvard Business School (180,000+ negotiations), agents extract private information through prompt injection. Limit agent authority in advance.

Source 1: https://www.ft.com/content/f7c26ac0-f5f2-443a-86b2-39d62f7920a9

Sources