Safeguarding Proprietary Data In The AI Era

By Seth Kugler

Having merely six months of background in a sector typically does not prepare someone to siphon off a former employer’s clients and launch a rival venture. Naturally, that perspective can change if the plan involves leveraging proprietary company details to train an artificial intelligence agent capable of doing the heavy lifting.

That exact scenario unfolded in a recent federal trade secret lawsuit litigated by Seth Kugler on behalf of a client in the U.S. District Court for the Northern District of California. In that litigation, a former senior marketing agency employee did not simply copy client directories to reach out to accounts. Instead, the individual utilized stolen trade secrets as training inputs to essentially construct an AI replica of his former boss, seeking a shortcut to establish a rival company without spending years building the requisite skills and expertise traditionally necessary to deliver those services.

Artificial intelligence is giving rise to novel methods of trade secret theft, extending far beyond the traditional scenario of an employee transferring files onto a USB drive. This dispute illustrates a new reality for startup founders, bringing to light unprecedented vulnerabilities concerning artificial intelligence, trade secrets, staff utilization of personal hardware, and the heightened necessity for businesses to safeguard themselves in the era of agentic AI.

Key lessons for founders

AI contamination cannot be undone

During the lawsuit, the defense asserted that because their client placed the purloined records into a designated quarantine directory and had not yet poached any customers, no damage had occurred.

The judicial system rightly determined that the quarantine folder was wholly inadequate, approving a temporary restraining order and a preliminary injunction while describing the rival AI agent as an existential threat.

A pledge to avoid touching the pilfered files directly loses its meaning once artificial intelligence enters the picture: When a trade secret helps develop a machine learning model or trains an AI agent featuring recursive capabilities, the model becomes contaminated, and undoing that consequence is exceedingly difficult. Even if raw source files are thoroughly purged of any direct evidence pointing to stolen content, defendants will face significant challenges convincing courts that the AI is not functioning a certain way precisely because it once accessed that proprietary information.

Personal devices can pose big risks

Historically, trade secret misappropriation was generally intentional, but artificial intelligence now introduces substantial dangers regarding accidental infractions. Startups routinely expand while personnel operate on personal hardware under standard Confidential Information and Invention Assignment Agreements. Running desktop AI applications on these systems can present a major hazard.

Unlike standard web-based interfaces, a desktop AI agent has the ability to scan documents, generate folders, and ingest directory paths across a local storage drive, frequently without demanding special authorization or notifying the user.

At present, thousands of workers might unwittingly breach non-disclosure agreements by operating consumer AI tools on personal laptops that store employer intellectual property. If the active configurations permit model training using this information, tribunals could eventually rule that these trade secrets forfeited legal protection due to dissemination outside the corporate entity.

Standard IT audits won’t catch AI data theft

This recent legal matter acts as a caution regarding how attorneys and founders must approach trade secret investigations. Conventional digital discovery is no longer adequate because AI technologies allow individuals to easily scatter, modify, and recreate data across nonobvious directory paths.

Measures that once seemed excessive—such as performing full forensic imaging on a personal device—now represent the bare minimum. Additionally, artificial intelligence dramatically lowers the barrier for reverse-engineering software and technology products, potentially eliminating obvious indicators that code or data was copied.

Demonstrating theft within artificial intelligence environments demands intensive, highly technical expert evaluations and entails far greater unpredictability than standard intellectual property litigation.

3 practical steps companies should take

Founders ought to execute three operational and legal measures to proactively protect their enterprise:

1. Implement strict agentic AI policies: Enforce rigorous protocols and settings that segregate external AI from company documentation. Browser-based platforms such as Claude, which feature strict data privacy rules, are significantly safer than open-source programs or desktop applications that crawl local drives. Certain AI agents should be barred entirely from running inside any environment housing trade secrets.
2. Eliminate remote access from unmanaged devices: Transition toward a secure cloud environment where data can be viewed and handled on a protected server, preventing files from being downloaded locally to a personal desktop. Restricting local file downloads removes the hazard of desktop AI tools indexing proprietary information.
3. Rewrite your Confidential Information and Invention Assignment Agreements for the AI era: Standard non-disclosure agreements require explicit clauses covering AI usage, model training inputs, and a precise description of data misappropriation via machine learning.

Should you suspect that an employee has departed with data or misused it inside an AI platform, your immediate action must be to cut off all system access and retain legal counsel. Do not wait until client attrition or financial losses begin to mount. Feeding trade secrets into artificial intelligence creates an existential danger from day one.

Artificial intelligence grants individuals the unprecedented capacity to compete within domains where they lack experience, rendering proprietary data more exposed than ever. Modernizing security, policy, and legal frameworks to tackle contemporary AI threats will assist startups and mature organizations in preventing catastrophic exposure.

---

Frequently Asked Questions

How does AI change traditional trade secret theft?

Rather than just copying files to a thumb drive or personal cloud account, individuals can feed stolen proprietary data into AI models to quickly train machine learning systems or build competing AI agents, effectively bypassing years of required experience.

Can a quarantined folder protect a company if stolen files are used in AI?

No. Courts have recognized that simply moving stolen documents to a quarantine folder is insufficient because AI models trained on that data become contaminated, and that impact is extremely difficult to undo.

Why are personal devices a major risk for trade secrets?

Desktop AI agents can scan local files, folders, and directory paths across a personal laptop’s hard drive without special permissions or user alerts, potentially exposing employer intellectual property to external model training.

How can companies protect their trade secrets from AI-related threats?

Companies should implement strict agentic AI policies, eliminate remote file downloads on unmanaged personal devices by moving to secure cloud environments, and update their Confidential Information and Invention Assignment Agreements to explicitly cover AI usage and machine learning data inputs.

What should a founder do if they suspect an employee has stolen data using AI?

The immediate step is to cut off all system access and retain legal counsel immediately before financial losses or client departures occur.

---

---

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.

Illustration: Dom Guzman