A recent federal trade secrets case in the U.S. District Court for the Northern District of California involved a former employee who used stolen proprietary data to train an AI agent. The defendant attempted to build a competing business by using the stolen information as inputs for a machine learning model.
The court granted a temporary restraining order and a preliminary injunction against the defendant. The judge characterized the competing AI agent as an existential threat, rejecting the defense's argument that moving stolen documents to a quarantine folder mitigated the harm.
The case highlights that once trade secrets are used to train AI models, the contamination is difficult to reverse. Even if raw files are removed, proving that an AI model does not rely on stolen proprietary information remains a significant legal challenge.
The use of personal devices for work poses risks, as desktop AI agents can scan local drives and ingest data without user permission. This can lead to inadvertent violations of confidentiality agreements and the potential loss of legal protection for trade secrets.
Legal experts recommend three steps for companies: implementing strict policies to isolate AI from sensitive documents, eliminating remote access to files from unmanaged devices, and updating confidentiality agreements to explicitly address AI use and model training.
