Court blocks AI agent trained on stolen trade secrets

The case, litigated by Seth Kugler of Grellas Shah LLP, involved a former senior employee at a marketing agency who did not simply copy client lists but used stolen trade secrets as training inputs to build an AI version of his old boss. This approach aimed to shortcut the years of experience typically needed to compete in that space. The court's recognition of the AI agent as an 'existential threat' suggests that judges may view the mere training of a model on proprietary data as harmful, even before any clients are poached. Kugler notes that traditional digital discovery is no longer sufficient because AI tools can disperse, alter, and recreate data across nonobvious folder paths, making full forensic imaging of personal devices the new bare minimum. He also points to inadvertent risks: thousands of employees may be violating confidentiality agreements by running consumer AI agents on personal laptops that house employer IP, and standard IT audits won't catch this. The three practical steps Kugler outlines are grounded in this new reality, emphasizing isolation of AI from company documents, closed cloud environments, and updated legal agreements.