Harvey: Extending Legal agent bench to M&A due diligence
Introducing a new extension to Legal Agent Bench that evaluates AI agents on one of legal’s most complex workflows: M&A due diligence.
Our goal with Legal Agent Bench (LAB) is to create and distribute realistic, high-scale agent environments to evaluate agents’ ability to perform end-to-end legal work and support open model training and agent research. Today, we’re extending LAB to cover one of the most critical legal tasks: M&A due diligence.
Diligence underlies every merger and acquisition which collectively amounted to around $4.8 trillion dollars of economic activity in 2025. Diligence costs typically range from 1-4% of deal value, landing them between $50 and $200 billion per year. Much of that cost is spent reviewing a virtual data room (VDR), systematically working through a company’s legal and financial history to identify risks, mitigate them, and confirm the deal lives up to expectations.
To develop LAB environments for diligence, we wanted to focus on emulating the depth and complexity of a real VDR. This requires building novel evaluative environments that scale both size and depth of evaluation.
To read the article in full, click here.
