Harvey tests AI agents on U.S. Case-Law research and brief writing. The new framework evaluates whether frontier models can review real case records, research precedent and draft litigation briefs, while Harvey warns that significant failure risks remain. Harvey has introduced an evaluation framework to test whether advanced artificial intelligence models can research U.S. case law and draft litigation briefs using real court records. The company says a small group of frontier models can produce complete first drafts in a single attempt, without attorney guidance, but cautions that the work still has a long tail of failure modes. The framework, LAB: United States Case Law (LAB: USCL), focuses on a demanding part of litigation: preparing written arguments for contested motions. Lawyers must understand the record, find relevant legal decisions and explain why those decisions support a requested court ruling. Harvey says the emerging model capability is a substantial advance, though its briefs are not presented as ready to file without review. The announcement matters because a brief can misstate the law if it relies on precedent that has since been narrowed or rejected. Courts generally must follow controlling decisions from higher courts in their jurisdiction, while decisions from other courts may be persuasive but are not binding. What the evaluation tests. LAB: USCL builds research environments from real, public data collected by Free Law Project and compares model-written briefs head-to-head. The task is designed to cover the steps a lawyer takes: examining the case docket, researching precedent and drafting an argument. Dockets can contain thousands of filings, including complaints, motions, exhibits and court rulings, making it necessary to understand how the record developed rather than rely on a single document. Case-law research also requires checking how later rulings affect earlier opinions. An older decision may appear to support a party's position in isolation, even though a higher court has since changed the governing rule. Harvey says LAB: USCL adds new approaches to its broader LAB work to examine model behavior on these tasks. What remains unclear. Harvey's announcement does not identify the models in the successful group, quantify their number or provide detailed results for individual briefs. It also does not say how often the systems made errors or specify which failure modes appeared most often. The company describes the drafts as substantially closer to the standard expected for court filings, while acknowledging that serious shortcomings remain. Harvey said it plans to make research in this area more accessible in partnership with Grubel. The announcement does not provide a timeline or explain what that access will involve. This report draws on information from Harvey. Editorial checks passed Facts, figures and quotes checked against 1 source

Grubel raises €3M pre-seed. Grubel raises €3 million in pre-seed funding led by Point Nine to automate the AI specialisation loop for complex legal work. Updated September 11, 2026 Grubel, a Munich- and Tübingen-based AI research lab developing systems that adapt to individual legal matters, has raised €3 million in pre-seed funding to support research, product development, and team expansion. Investors. The round was led by Point Nine, with participation from angel investors including Jeff Dean, Chris Ré, Ion Stoica, Gabe Pereyra, and Winston Weinberg. Grubel use of funds. Grubel plans to use the capital to support its research, product development, and team expansion efforts. About Grubel. Founded in 2026 by Moritz Hardt and Reinhard Heckel, Grubel is an AI research lab focused on automating the process of tailoring AI systems for specific legal workflows. The company is developing a specialisation loop that gathers and prepares task-relevant information, adapts AI models to specific legal matters, and evaluates output against case-specific standards. Funding details. Company: Grubel Raised: €3M Round: Pre-seed Funding Date: September 10, 2026 Lead Investor: Point Nine Additional Investors: Jeff Dean, Chris Ré, Ion Stoica, Gabe Pereyra, Winston Weinberg Software Category: Legal Tech / AI Source: https://www.eu-startups.com/2026/09/new-german-ai-research-lab-grubel-raises-e3-million-to-build-systems-that-adapt-to-individual-legal-cases/ Updated September 11, 2026