Anthropic Pilots AI Agents for Legal Transactions in Project Deal Test
Anthropic's Project Deal, conducted in May 2024, saw AI agents autonomously execute 186 legal-style transactions.
Why it matters: AI that conducts basic transactions could cut billable hours for lawyers and shift workflows for legal teams. Early results offer a glimpse of AI’s impact on legal operations and transactional work.
- Project Deal ran in May 2024, with 69 employees receiving $100 each to use in an AI-managed barter marketplace.
- AI agents autonomously negotiated and completed 186 transactions totaling over $4,000 in value.
- Anthropic’s advanced Claude Opus model outperformed Haiku, achieving a $3.64 higher average sale price.
- Global firm Freshfields adopted Anthropic’s AI, driving 500% growth in usage among 5,700 staff within six weeks.
Anthropic’s Project Deal (May 2024) gave 69 staffers a $100 budget each for a controlled internal marketplace involving over 500 physical or digital items. Participants delegated negotiations and closing to AI agents, simulating tasks commonly handled by in-house counsel or transactional associates.
- Agents managed 186 total transactions raising a combined $4,000+ without human intervention on deal terms or execution.
- The company compared AI agent performance, finding that the Claude Opus model resulted in a $3.64 higher average sale price than the simpler Haiku version.
This proof-of-concept aims to demonstrate how large language models might automate routine legal or commercial agreements. For legal teams and firms, this could streamline high-volume, low-complexity work, potentially lowering costs and freeing up human lawyers for higher value tasks.
Major firms are watching. At Freshfields Bruckhaus Deringer, a global law firm, a partnership with Anthropic enabled rapid AI adoption: within six weeks of introducing Claude, staff AI use increased fivefold across 33 offices. Gil Perez, who leads the Freshfields project, said the initiative enables "co-innovation at pace" and the secure integration of AI for legal needs.
Legal industry analysis by Legaltech News notes legal AI agents excel in automating simple, repetitive transactions but face challenges in explaining decisions, ensuring compliance, and managing errors—key risks for in-house counsel and compliance teams. Experts urge caution: "Black-box negotiation presents real oversight risks for corporations," said a legal operations analyst at Legaltech News.
Project Deal focused on straightforward barter arrangements. Broader adoption for complex or regulated deals will require robust explainability, compliance controls, and human-in-the-loop review to satisfy professional and regulatory standards.
By the numbers:
- 186 — number of transactions AI agents completed autonomously in Project Deal.
- $4,000+ — total value of barter marketplace transactions managed by AI.
- 500% — jump in AI usage at Freshfields among 5,700 staff after six-week introduction.
Yes, but: Project Deal only tested simple transactions; scaling AI agents to complex matters faces technical and compliance hurdles.
What's next: Law firms and in-house teams are piloting similar AI agent tools, with further real-world results expected later in 2024.