Anthropic is hiring lawyers to teach AI agents how to negotiate deals
The future of commerce is your AI haggling with their AI. Anthropic is paying lawyers up to $167,000 to write the rules of engagement.
Models · Source: The Decoder
What happened
Anthropic is consulting law firms to figure out how AI agents should negotiate on behalf of users. According to reports, the goal is to set fair norms before these agents start haggling over prices, resolving complaints, or pushing for better salary offers. The AI company is deeply studying how human lawyers advocate for their clients. They want to determine exactly how much freedom AI agents should have when making decisions with limited human involvement.
This push comes from a direct warning sign in their own research. Anthropic ran experiments showing that stronger AI models negotiate significantly better deals than weaker ones. They realized this performance gap could widen societal inequality. If rich users get aggressive, top-tier AI negotiators while others rely on weaker models that silently lose out, the economic divide worsens.
To fix this data problem, Anthropic is hiring Conceptual Reasoning Fellows. They are specifically looking for lawyers. These fellows will spend four months testing Claude on complex, judgment-heavy questions across law, philosophy, cognitive science, economics, and political science. The company is paying these experts between $100,000 and $167,000 to build test cases. They need experts because right and wrong answers in the humanities are rarely clear-cut.
Key facts
- $100,000 to $167,000 — Pay for a four-month Conceptual Reasoning Fellowship at Anthropic
- Four months — Duration lawyers will spend testing Claude on judgment-heavy questions
Why it matters
We are moving past AI that just writes code or drafts emails. Agents are starting to make financial decisions and secure contracts. If an AI is negotiating your next salary, the definition of a good deal becomes a technical benchmark. Justin Curl, a member of the technical staff researching law and institutions at Anthropic, pointed out that quality in law is notoriously hard to define. Builders now have to figure out how to evaluate an AI model on highly subjective human interactions.
The second-order effect is a massive shift in how software interacts with the economy. Improving AI models in math and code is relatively easy because verifiable data is plentiful and easy to generate. Training them on human negotiation requires entirely new datasets based on legal and ethical boundaries. If Anthropic cracks this, the entire landscape of digital commerce changes. You will not be negotiating with a vendor. Your agent will negotiate with their agent to find the optimal contract.
For builders
Building subjective evaluation datasets
Math and code have clear answers but humanities and law do not. Founders who can build verifiable evaluation tools for subjective AI tasks will find massive demand from foundation model companies. Anthropic is paying lawyers top dollar because this data does not exist yet.
Agent to agent commerce infrastructure
Soon AI agents will haggle over prices and contracts autonomously. Engineers need to build the infrastructure, APIs, and guardrails that allow two AI models to securely transact. The companies that build the payment and contract rails for agentic commerce will capture immense value.
Translating legal constraints into code
Anthropic is turning legal norms into model behavior to prevent societal inequality. Startups that can translate complex legal and ethical boundaries into system prompts or fine-tuning data will win major enterprise contracts. Businesses will pay heavily to ensure their negotiating agents do not break the law.
My take
The future of shopping is not you against a salesperson. It is your AI against their AI, and both are spending your money. Anthropic hiring lawyers proves that the next frontier for models is not just reasoning, but ruthless and optimized haggling. If you are building consumer tech, prepare for a world where the user never actually negotiates a deal themselves.
Original reporting: The Decoder. This is my rewrite and opinion.