Harvey Expands Legal Agent Bench to M&A Due Diligence
Harvey, the AI-focused legal tech company, has announced an expansion to its open-source Legal Agent Bench (LAB) to tackle one of the most demanding legal workflows: M&A due diligence. This extension aims to evaluate AI agents' ability to manage the intricate, high-stakes processes involved in mergers and acquisitions, a market that saw $4.8 trillion in activity in 2025, according to Bain & Company.
Due diligence is a cornerstone of M&A transactions, with costs typically ranging from 1-4% of deal value, equating to $50-$200 billion annually. These expenses often stem from labor-intensive reviews of virtual data rooms (VDRs), where hundreds or even thousands of documents outlining a company's legal and financial history are scrutinized. Harvey's LAB now introduces synthetic VDR environments to test AI agents' ability to navigate these complexities, offering a scalable, cost-efficient alternative to traditional human-led processes.
How LAB Tackles Due Diligence
LAB’s new diligence environments simulate the depth and breadth of real-world VDRs. For example, one synthetic case involves the hypothetical acquisition of Sentinel Cloud Security by Helios Cloud Holdings, modeled after Google’s $32 billion acquisition of Wiz. These environments include over 3,500 documents and 45 million tokens of context, requiring AI agents to identify risks ranging from intellectual property disputes to tax liabilities. Agents are evaluated on their ability to draft diligence memoranda, with their output graded against hundreds of rubric criteria.
This approach addresses critical challenges in AI handling of legal workflows, such as managing vast quantities of context, performing exhaustive reviews, and applying judgment to prioritize risks based on materiality. Unlike existing AI models that rely on keyword searches or selective document reading, Harvey’s agents aim to comprehensively review all data while maintaining the ability to synthesize complex issues across multiple files.
Why This Matters
M&A due diligence is not just about fact-finding; it’s about assessing the value and risks of a deal. Effective diligence shapes deal terms, including pricing, warranties, and post-closing conditions. By automating portions of this process, AI agents could drastically reduce costs and timelines, critical factors in a market where deals often hinge on rapid execution.
Harvey’s LAB stands apart from earlier benchmarks, such as LegalBench, by focusing on multi-step workflows rather than isolated reasoning tasks. This aligns with Harvey's broader strategy of developing agentic legal systems that move beyond simple Q&A functionalities to handling end-to-end legal matters. The company’s May 2026 rollout of purpose-built legal agents underscores this shift, aiming to enable law firms to measure AI’s economic impact and improve ROI.
What’s Next?
Harvey plans to release research in the coming weeks detailing strategies for training diligence agents and their performance across diverse synthetic VDRs. The company is also working to transition these benchmarks from research to production, offering law firms a data-safe environment to train and refine custom models. Additional LAB extensions are expected to cover tasks such as enterprise search and fund formation, broadening the scope of AI applications in legal practice.
As the legal industry increasingly adopts AI-driven tools, Harvey’s LAB could redefine how firms approach high-cost, labor-intensive workflows like M&A due diligence. While challenges remain in training agents to handle multi-document reasoning and nuanced judgment, the potential to streamline multi-billion-dollar transactions is drawing significant attention from firms and tech providers alike.