Grubel Raises €3 Million to Build AI Systems Tailored to Individual Legal Cases

German AI research lab Grubel has raised €3 million in pre-Seed funding to develop a new approach to legal artificial intelligence—one designed to adapt specifically to the facts, documents, terminology, and requirements of each individual legal matter. The company plans to use the fresh capital to accelerate research, develop its product, and expand its team.

Funding Round Led by Point Nine

The funding round was led by European venture capital firm Point Nine and attracted a strong group of prominent technology and AI researchers as angel investors.

Participants include Jeff Dean, former Chief Scientist at Google; Chris Ré, Stanford professor and co-founder of Together AI; Ion Stoica, UC Berkeley professor and co-founder of Databricks; and Gabe Pereyra and Winston Weinberg, co-founders of legal AI company Harvey, among other investors.

Point Nine Partner Louis Coppey said the team is taking an ambitious approach to AI research in the legal sector and believes its technology could eventually unlock new capabilities across other areas of complex knowledge work.

Founded by Leading Machine-Learning Researchers

Founded in 2026, Grubel was created by machine-learning researchers Moritz Hardt and Reinhard Heckel, bringing together deep academic and research experience in artificial intelligence.

Moritz Hardt

Moritz Hardt is a Director at the Max Planck Institute for Intelligent Systems and previously served as a professor at UC Berkeley and worked as a researcher at Google Brain. His research includes test-time training, and he has also contributed to Lawma, a research project focused on language models and legal applications.

Reinhard Heckel

Reinhard Heckel is a professor of machine learning at Technical University of Munich (TUM) and is currently on leave. Before joining academia, he worked as a researcher at IBM Research. His work focuses heavily on data-centric machine learning, and he has contributed to projects including DataComp-LM and OpenThoughts.

Why Legal AI Needs a Different Approach

While AI has made major advances in areas such as coding and general information retrieval, Grubel believes that highly specialised knowledge work remains difficult to automate.

Legal work is a particularly challenging example. Every case can come with its own facts, documents, terminology, previous work, client expectations, and standards for what constitutes a good result. This means a general-purpose AI model may not always have the right context or evaluation framework to handle a complex legal matter effectively.

Today, specialised legal AI systems often require lawyers and engineers to manually collect relevant information, prepare datasets, adapt models, and evaluate their performance. According to Grubel, this approach can work, but it is difficult to scale when every legal matter requires a different system.

Grubel's Vision: Every Legal Matter Needs Its Own AI

Grubel is taking a different approach. Its central idea is that each complex legal matter should have an AI system specifically adapted to that matter.

Rather than building one fixed AI system and expecting it to work across every case, Grubel is developing an automated AI specialisation loop that continuously adapts the system to the work it needs to perform.

The company's technology is built around three core components:

1. Data Engine

The system identifies and curates the information that is relevant to a particular legal matter or client. This can include documents, previous work, terminology, and other task-specific data.

2. Test-Time Adaptation

The technology then adapts both the AI model and the agent to the specific matter before it begins carrying out the task.

3. Continuous Evaluation

The system evaluates its own work against standards derived from the individual matter, allowing it to identify where improvements are needed and repeat the process.

Together, these elements create a continuous cycle of data curation, adaptation, and evaluation, allowing the AI system to become increasingly specialised for the work in front of it.

Early Results Show Promise

Grubel says an initial evaluation found that its specialisation approach was able to outperform general-purpose frontier AI systems on the legal tasks tested.

The company now plans to further improve the technology so that it can handle increasingly complex legal work, while simpler and more routine tasks can potentially be handled by smaller, more efficient AI models.

Funding to Accelerate Research and Product Development

The newly raised €3 million will be used across three major areas: AI research, product development, and team expansion.

The company aims to turn its research-driven approach into a scalable platform capable of automatically creating specialised AI systems for individual legal matters. In the longer term, Grubel believes the same underlying technology could extend beyond legal services to other forms of complex knowledge work.

A New Direction for Legal AI

Grubel's funding comes at a time when legal technology is increasingly exploring how AI can move beyond general-purpose assistants and become more deeply integrated into professional workflows.

Instead of asking whether one AI model can handle every legal problem, Grubel is exploring a different question: Can AI adapt itself to the specific problem it is being asked to solve?

With €3 million in fresh backing and a founding team with deep expertise in machine learning, the Munich- and Tübingen-based lab is positioning itself around that idea—building AI systems that learn what a particular legal matter requires before attempting to solve it.