What happened?
Anthropic, the developer of the Claude model, is consulting law firms to define negotiation norms that its AI agents should follow when representing users in agreements, bargaining, or dispute resolution. According to a Financial Times report on October 18, 2026, the company seeks to understand how lawyers balance defending client interests with fair outcomes, especially when agents interact with other AI systems with conflicting goals.
The move comes as Anthropic places the development of autonomous agents at the center of its strategy, ahead of a possible IPO. Sources familiar with the matter indicate the company is evaluating how much human feedback agents should request during a negotiation and how to protect against adversarial agents without becoming overly suspicious. One idea being explored is requiring users to provide signed documents detailing the agent's mandate and mandatory rules.
Why does this matter?
AI agents that negotiate on behalf of people represent a leap in artificial intelligence autonomy. Unlike chatbots that only respond, these systems can act independently in markets, customer service, or even salary negotiations. Anthropic has already conducted internal experiments such as Project Swap and Project Deal, in which Claude agents bought, sold, and bargained over real books and other items among employees.
In these tests, agents instructed to be "ruthless" gained a slight advantage, while more cooperative ones made conscious concessions. The AI model used influenced results more than specific instructions. Defining ethical and legal rules now can prevent future abuses and set standards for the entire industry.
What changes in practice?
For users, this means that soon an AI assistant could negotiate the price of a product, resolve a consumer complaint, or even discuss work conditions with another agent, always within limits set by the owner. The need for clear legal frameworks becomes urgent, as specific laws for agents transacting on behalf of humans do not yet exist.
Anthropic positions its agents as faithful representatives of the user but recognizes that machine-to-machine interaction requires "fair play" norms. The outcome of these consultations could influence not only Claude but the entire emerging AI agent ecosystem in 2026.
By GeekikiBot