If you only have six months of experience in an industry, you usually aren’t equipped to poach your employer’s clients and start a competing business. Of course, you may believe otherwise if you plan to use your employer’s proprietary information to train an AI agent to do the heavy lifting for you.
That is exactly what the defendant attempted in a recent federal trade secrets case I litigated on behalf of a client in the U.S. District Court for the Northern District of California. In that case, the defendant, a former senior employee at the marketing agency, didn’t just copy client lists to solicit accounts. Instead, he used stolen trade secrets as training inputs to build, essentially, an AI version of his old boss, attempting a shortcut to create a competing business without spending years developing the skills and expertise typically needed to provide such services.
AI is creating new ways to misappropriate trade secrets, well beyond the classic theft story of an employee downloading files onto a thumb drive. This case highlights a new reality for founders, exposing unprecedented risks related to AI and trade secrets, employee use of personal devices, and the heightened need for companies to protect themselves in the age of agentic AI.
Key lessons for founders
AI contamination cannot be undone
The defense in this case argued that because their client placed the stolen documents in a “quarantine folder” and no clients had been poached yet, no harm had been done.
The court appropriately recognized that the “quarantine folder” was grossly insufficient, granting our temporary restraining order and preliminary injunction while recognizing the competing AI agent as an “existential threat.”
An agreement not to touch just the stolen documents themselves is essentially irrelevant once AI gets involved: If a trade secret is used to develop a machine learning model or used to train an AI agent with recursive capabilities, it contaminates the model and the impact can be difficult to undo. Even if raw source files are completely scrubbed of any direct evidence of stolen materials, defendants will struggle to prove to courts that the AI isn’t operating a certain way precisely because it once had access to that proprietary information.
Personal devices can pose big risks
While trade secret misappropriation has typically been malicious, AI now also creates significant risks of inadvertent violations. Startups frequently scale with employees working on personal devices under standard Confidential Information and Invention Assignment Agreements. Running desktop AI agents on these devices can pose a serious risk.
Unlike standard web interfaces, a desktop AI agent can scan files, create folders and ingest directory paths across a local hard drive, often without requiring special permission or alerting the user.
Right now, thousands of employees may be inadvertently violating confidentiality agreements by running consumer AI agents on personal laptops that house their employers’ IP. If settings allow model training from this data, courts may eventually find these trade secrets to have lost legal protection due to dissemination outside the company.
Standard IT audits won’t catch AI data theft
This recent case serves as a warning for how attorneys and founders must approach trade secret investigations. Traditional digital discovery is no longer sufficient because AI tools make it effortless to disperse, alter and recreate data across nonobvious folder paths.
Actions that previously seemed like overkill — such as full forensic imaging of a personal device — are now the bare minimum. Furthermore, AI dramatically reduces the friction of reverse-engineering software and tech products, potentially removing telltale signs that code or data was copied.
Proving theft in AI contexts requires intensive, highly technical expert analysis and carries far more uncertainty than traditional IP litigation.
3 practical steps companies should take
Founders should take three legal and operational steps to proactively protect their business:
- Implement strict agentic AI policies: Mandate tight controls and settings that isolate outside AI from company documents. Web-based tools like Claude with strict data privacy policies are vastly safer than open source or desktop apps that crawl local drives. Some AI agents should not be permitted to run at all within an environment containing trade secrets.
- Eliminate remote access from unmanaged devices: Shift to a closed cloud environment where data can be viewed and manipulated on a secure server, but files cannot be downloaded locally to a personal desktop. Removing local file downloads eliminates the risk of desktop AI agents indexing proprietary data.
- Rewrite your Confidential Information and Invention Assignment Agreements for the AI era: Standard confidentiality agreements need explicit clauses addressing AI use, model training inputs, and a precise definition of data misappropriation through machine learning.
If you suspect an employee has walked away with data or misused it in an AI tool, your first step should be immediate: cut off all system access and retain counsel. Don’t wait until you start losing clients or experiencing financial losses. Feeding trade secrets into AI presents an existential threat from day one.
AI gives individuals the unprecedented ability to compete in spaces where they lack experience, making proprietary data more vulnerable than ever. Modernizing security, policy and legal strategies to address today’s AI threats will help startups and established companies avoid catastrophic exposure.
Seth Kugler is a partner at Grellas Shah LLP, where he focuses on complex commercial litigation, including disputes related to trade secret and IP misappropriation, corporate governance and breach of fiduciary duties, breach of contract, and other complex business disputes. He specializes in preventing and resolving all manner of disputes among and between corporate founders, directors, officers and shareholders, LLC members, and other current and former business partners.
