Legal AI: Former Meta Employee Aims for $6M Raise

Firms like Kirkland & Ellis and Latham & Watkins are funding proprietary AI tools, including the purchase of Nvidia servers. In this context, Ethem Coban, who previously worked at Meta's Superintelligence Labs, left the company this month to create his own startup and is seeking to raise funds.

Firms Invest $500M and Acquire Nvidia Servers

Kirkland & Ellis, which ranks first in the industry in terms of revenue, has invested $500 million in developing its own software. Latham & Watkins, on the other hand, has acquired Nvidia servers to customize models in-house. Firms see their vast amounts of legal documents as a source of competitive advantage that others cannot simply buy.

Ethem Coban: Two Years at Meta in Superintelligence Labs

Ethem Coban spent nearly two years at Meta, primarily within the Superintelligence Labs. He contributed to training models designed to better answer legal questions and reduce hallucinations. His work also focused on teaching legal reasoning to language models. Before entering the industry, he studied law at American University and Goethe University Frankfurt.

Goal: Custom Models and a Targeted Fundraising of $6M

Coban left Meta this month to found a startup specializing in legal technology. According to a source familiar with the matter, he is currently discussing a fundraising round of $6 million, based on a valuation of $25 million; these negotiations, still in a preliminary stage, could see their terms evolve. He has informed investors that he aims to assist firms and legal teams in creating and adapting models based on their expertise. He argues that companies will increasingly want to own their AI rather than lease it, in order to maintain control over their data. The market is attracting talent from major labs and already has numerous players, with pioneers like Harvey and Legora benefiting from strong demand, while lawyers have been experimenting with these tools for two years. In his view, the challenge will be less about convincing the need for AI than demonstrating that an intimate understanding of how models work is crucial for designing them.