Anthropic Turns to Legal Experts to Train AI Agents in Negotiation as Autonomous Systems Prepare to Strike Deals

Anthropic Partners with Legal Firms to Define AI Negotiation Standards

Anthropic is consulting with legal experts to solve a pressing challenge in artificial intelligence: how AI agents should negotiate deals on behalf of their human users1

1

. The San Francisco-based company is seeking advice from law firms to understand how lawyers interact during negotiations while representing clients and honoring fair outcomes1

1

. This initiative comes as Anthropic positions the development of AI agents at the core of its strategy ahead of a planned initial public offering that investors believe could become the biggest stock market listing in history1

1

.

The company is exploring negotiation norms for agentic systems as these autonomous agents increasingly represent users while encountering other agents with conflicting goals1

1

2

2

. Sources familiar with Anthropic's thinking reveal the company is balancing critical factors: how much human feedback agents should seek during negotiations and how suspicious agents should be of AI counterparts without becoming paranoid or conspiratorial1

1

2

2

.

Why AI Agents Strike Deals Matters for Users and Businesses

The rise of personal agents is forcing tech companies to confront how machines will communicate with one another and negotiate on behalf of humans1

1

. These autonomous agents could haggle over product prices, resolve customer service disputes, or even secure better salary offers1

1

. Setting rules of engagement for agents has become increasingly critical as the technology sees wider adoption. Meta has experienced a significant jump in uptake of its personal agent tool Muse, while OpenAI recently released a competitor product called Dots1

1

.

Anthropic has explored giving agents signed documents from users verifying their brief and any rules to follow1

1

. The company acknowledged in research posted in April that policy and legal frameworks around AI models that transact on behalf of humans simply don't exist yet, warning that such a world isn't far away and society will need to move quickly to address these changes1

1

.

Real-World Experiments Reveal Negotiation Tactics and Challenges

Anthropic has published findings from multiple experiments this year where its Claude agents negotiated over real goods and money1

1

. In September, Anthropic employees engaged in an experiment to buy and sell books using AI agents on a virtual trading floor to pitch, haggle, and strike deals with other people's agents1

1

. Employees gave their agents a list of books ranked by preference to teach AI agents to negotiate effectively.

The study found that agents instructed to be ruthless came out slightly ahead of other agents in securing better outcomes1

1

. More cooperative agents made knowing sacrifices, including accepting less-desired books to ensure each agent ended up with a book to return to the user1

1

. These experiments demonstrate how negotiation skills and strategies vary based on agent programming and objectives.

Academic Research Uncovers Novel Tactics and Inequality Concerns

Michelle Vaccaro, an assistant professor at Harvard Business School, ran more than 180,000 negotiations between AI agents across scenarios including employment contracts and rental contract bargaining1

1

. Her research identified novel tactics that undermine another agent's ability to represent the user. In one case, an agent tried to coax a counterpart into revealing private information through a technique called prompt injection1

1

.

Vaccaro highlighted concerns about bargaining power inequality. Highly capable models could make negotiation expertise accessible to people who previously lacked the time or resources to develop those skills themselves or hire someone to negotiate on their behalf1

1

. However, if the strongest agents are available primarily to those with the most resources, they could reinforce existing advantages and widen inequalities in bargaining power1

1

.

Broader Applications Beyond Consumer Transactions

The implications of agentic AI extend beyond consumer negotiations into enterprise applications. In cross-border treasury, the technology's strongest early applications include automated beneficiary-data validation to reduce avoidable rejections and exception-management agents to identify recurring issues by bank, supplier, or payment corridor2

2

. Long-term opportunities could see agents coordinate complex payment workflows across providers as banks improve application programming interfaces, harmonize messaging, and extend settlement availability2

2

.