Artificial intelligence company Anthropic is reportedly consulting law firms to understand how AI agents can negotiate deals on behalf of users, including bargaining over prices, resolving disputes and securing better salary offers.
The company is examining how much independence these agents should have while ensuring they operate within legal and ethical boundaries.
How Could AI Agents Negotiate Deals?
AI agents could negotiate laptop prices, rental agreements, customer complaints and employment contracts by communicating directly with sellers’ or employers’ AI systems.
However, when agents represent parties with conflicting interests, negotiations could become complicated. Anthropic is exploring “negotiation norms” and signed user instructions to define what agents can do independently and when they must seek human approval.
In September, Anthropic tested AI agents in simulated book-trading exercises. Agents instructed to negotiate ruthlessly performed slightly better at securing favourable deals, while cooperative agents sometimes accepted less-preferred books to ensure everyone received something.
Research by Harvard Business School assistant professor Michelle Vaccaro, involving more than 180,000 AI-agent negotiations over scenarios such as job offers and rental contracts, also identified security risks.
Agents could use prompt injection to manipulate others into revealing private information.
The technology raises concerns about inequality, as wealthier individuals and businesses with access to stronger AI models could gain negotiating advantages.
With legal frameworks for autonomous transactions still developing, clear authorisation, confidentiality safeguards and human oversight will be crucial to ensuring agents protect users’ interests without exceeding their authority.
