Intellectual Property & Technology Blog

AI already cost your company one form of protection over employee work product. Weak governance can cost you the second.

In January 2026, a federal court in the Northern District of California threw out a trade secret lawsuit against OpenAI. The plaintiff, Rebecca Trinidad, claimed she had developed proprietary frameworks for “emergent identity” and “autonomous multi-agent collaboration,” and that OpenAI had misappropriated them. The court didn’t need to decide whether her ideas had value. It dismissed her Defend Trade Secrets Act claim with prejudice for a simpler reason: she had built those frameworks by talking to ChatGPT, which meant she had voluntarily handed them to OpenAI the moment she created them. You cannot misappropriate a secret the plaintiff already gave you.

I read Trinidad as a governance failure, not a legal fluke. Most companies already have one hand tied behind their back when it comes to protecting AI-generated work product. Bad AI governance is what ties the other one.

The Tool You’ve Already Lost

Start with what copyright can no longer do for you, no matter how well you govern AI use. The Copyright Office’s January 2025 report and the D.C. Circuit’s decision in Thaler v. Perlmutter, left standing when the Supreme Court denied certiorari this March, confirm that AI cannot be an author and that prompting one does not make a human author either. Copyright protects only the parts of a work a human meaningfully shaped. If your employees generate reports, code, or analysis largely through AI prompts, a meaningful share of that output is not copyrightable at all. No policy fixes this. It is a structural limit on the tool itself, not a governance gap.

The Tool You Can Still Lose

Trade secret law doesn’t share copyright’s blind spot. The DTSA and its state counterparts ask only two questions: does the information have value because it’s secret, and did the owner take reasonable measures to keep it that way? Nobody asks who, or what, wrote it. That makes trade secret law the one protection still available for a large share of AI-generated work. It is also the one bad governance can still destroy.

Trinidad lost not because the law failed her, but because she satisfied neither element. She built her “secret” inside a tool operated by a third party, under that party’s terms of service, with no confidentiality obligation running the other way. That is the fact pattern ungoverned AI use creates every day, one prompt at a time. In United States v. Heppner, Judge Rakoff of the Southern District of New York ruled similarly on privilege: documents a defendant created using Claude weren’t protected because Anthropic’s terms of service allowed it to log and use the inputs. The court treated the vendor’s terms as proof the user had no real expectation of confidentiality. That logic travels easily from privilege to trade secrets.

What AI Governance Actually Requires

Governance here does not mean a slide deck employees skim once. It means routing sensitive work to enterprise AI tools with contractual data isolation and no training rights, not free consumer chatbots. It means confidentiality and NDA provisions that cover what employees put into AI tools, not just what they put in emails. And it means treating IP assignment language honestly: it now covers only the human-shaped slice of AI-assisted work, not the AI-generated substrate underneath it.

Copyright’s limits on AI-generated work are not a governance failure. They are the law. Trade secret protection is different. It is still available, and companies are losing it anyway, one ungoverned prompt at a time. Losing the tool the law already denied you is bad luck. Losing the one it left you is a choice.

Disclaimer: This blog does not constitute legal advice, does not create an attorney-client relationship, and is intended for informational purposes only.