Leadership and readiness
Can sharing trade secrets with a public AI tool waive protection?
By SourceX Editorial · Reviewed by Noah Loul ·
Short answer
Sharing trade secrets with a public AI tool can weaken or end their protection, because trade secret law requires reasonable measures to keep information secret. Entering it into a consumer AI service whose terms impose no confidentiality duty is the kind of disclosure courts scrutinize. The working rule: confidential material goes only into tools bound by written confidentiality terms.
Key takeaways
- Under federal law, information is a trade secret only if its owner takes reasonable measures to keep it secret.
- A disclosure to a recipient with no duty of confidentiality undercuts those measures, and a consumer AI account can be that recipient.
- Consumer and enterprise tiers of the same AI product can carry very different confidentiality, retention and training terms.
- A written AI use policy, approved accounts, enforced settings and documented training are evidence of reasonable measures.
- Records a company may later license should be screened for trade secret content and past AI tool disclosures before rights review closes.
Can sharing a trade secret with a public AI tool end its protection?#
Sharing a trade secret with a public AI tool can end its protection, though not automatically. Under 18 U.S.C. 1839(3), information qualifies as a trade secret only if its owner has taken reasonable measures to keep it secret and it derives independent economic value from not being generally known. A disclosure that the owner did nothing to control can be used to argue that neither condition still holds.
The principle is old. Courts have long treated handing information to someone with no obligation to keep it confidential as inconsistent with keeping it secret. What is new is the channel: employees now paste pricing logic, code and customer terms into chat tools many times a day without thinking of it as a disclosure at all.
Disputes that raise AI tool disclosures are now reaching courts, and commentators have discussed early decisions. Outcomes turn on the facts of each record and the terms of the specific tool, and no single trial-court decision settles the question everywhere. Ask counsel which current decisions apply in your jurisdiction before relying on any summary, including this one.
Why the AI tool's terms matter so much#
The AI tool's terms matter because they decide whether the provider is a confidential recipient or, legally, just another outsider. The federal Defend Trade Secrets Act, enacted in 2016, and the state laws modeled on the Uniform Trade Secrets Act both build in the reasonable-measures requirement.
DOJ guidance describes protective measures as needing to be reasonable under the circumstances rather than absolute, with examples such as telling employees the information is secret, limiting access on a need-to-know basis and requiring confidentiality agreements. A consumer AI account usually fails the last of those. Many AI products are sold in more than one tier, and consumer and business accounts of the same product can differ on confidentiality, prompt retention, human review and use of inputs for training. Check the current terms for the exact plan your staff use, because vendors revise them.
| Question to check | Why it matters for secrecy | Where to find the answer |
|---|---|---|
| Does the provider owe a duty of confidentiality for inputs and outputs? | Without that duty, the disclosure can look like a release to an outsider | Terms of service, enterprise agreement or data processing addendum |
| Can the provider use inputs to train or improve models? | Training use carries the information beyond the session and the account | Usage policy, privacy notice and admin training settings |
| How long are prompts, files and outputs retained? | Long retention widens the window for exposure and for discovery requests | Data retention terms and the admin console |
| Can provider staff review conversations? | Human review is a further disclosure to people outside the company | Privacy notice and abuse-monitoring terms |
| Who controls the account? | A personal account sits outside company policy, logging and deletion | Sign-up records and single sign-on configuration |
What the risk does and does not mean#
The risk means that pasting confidential material into an unapproved AI tool is a recognized way to weaken a trade secret claim; it does not mean every use of AI destroys protection. The difference lies in the tool's terms and in what else the company did to guard the information.
Most companies sit in the middle: some staff on approved enterprise accounts, others on personal free accounts nobody inventoried. That gap is what the checklist further down is meant to close.
- It does not make enterprise AI tools with written confidentiality and no-training terms off-limits for confidential work.
- It does not mean one employee's mistake automatically ends protection; courts weigh the owner's overall measures and the scope of the disclosure.
- It does not change what qualifies as a trade secret: the information must still derive value from not being generally known.
- It does mean opposing counsel may ask in discovery which AI tools held the information and under which terms.
- It does mean an AI use policy nobody enforces may be weak evidence of reasonable measures.
Which operational records are most exposed?#
The operational records most exposed are the ones staff paste to get quick help: pricing logic, estimating rules, source code, process parameters and customer-specific terms. They rarely look like secrets to the person copying them, which is why they leak.
In a software company, an engineer pastes a proprietary ranking function to debug it. In an engineering firm, a project manager drops a fee proposal with the firm's staffing assumptions into a chatbot to tighten the prose. In a distributor, a buyer uploads a supplier rebate schedule for a summary. In a manufacturer, a quality engineer pastes a root-cause analysis that names a proprietary process setting.
Each task is ordinary and well meant. The exposure comes from the account type, not the task: the same request on an approved account with confidentiality terms is a very different legal event from the same request on a personal free account.
A six-point reasonable-measures checklist for AI tools#
A reasonable-measures checklist for AI tools turns the legal requirement into controls a court can see: written rules, approved accounts, enforced settings, training and records. Each item should leave a document behind, because evidence of the measures matters as much as the measures.
When you update employee or contractor confidentiality agreements to cover AI tools, include the whistleblower immunity notice required by 18 U.S.C. 1833(b). An employer that omits it from contracts entered into or updated after the DTSA's enactment may lose the right to exemplary damages and attorney fees against that employee under the DTSA.
- Written policy: name approved tools, ban confidential material in unapproved ones, and define confidential in terms staff recognize, such as pricing, code, customer terms and process data.
- Approved accounts: company-managed enterprise accounts with confidentiality and no-training terms, tied to single sign-on so access ends when someone leaves.
- Enforced settings: training and history settings configured centrally where the product allows, with dated screenshots or exports kept as evidence.
- Access limits: network or browser controls that restrict consumer AI sites for the roles that handle the most sensitive material.
- Training and acknowledgment: short role-specific sessions, with signed acknowledgments from engineers, estimators, finance and sales operations.
- Incident response: a way to report an accidental disclosure, steps to delete the content where the tool allows, and a record of what was disclosed and when.
Illustrative: a machine shop finds its quoting logic in a chatbot history#
Illustrative: a fictional precision machining company prices work through its ERP and a spreadsheet of cost rules refined over many years. During a routine access review, IT finds that an estimator had pasted the spreadsheet's formulas into a personal account on a consumer AI tool to clean up the logic.
The general counsel treats it as an incident. The estimator deletes the conversations, IT captures the tool's terms as they stood at the time, and counsel records which rules were disclosed. The company moves estimators and engineers to an enterprise account with confidentiality and no-training terms, restricts consumer AI sites on estimating machines and adds AI tools to its confidentiality training.
The company neither assumes the cost rules are lost nor assumes they are safe. Counsel notes the disclosure in its trade secret register, so the issue is known before any dispute, sale or license rather than discovered during one.
Why AI tool disclosures matter before you license records#
AI tool disclosures matter before licensing records because a data license is the opposite of a consumer AI disclosure: a written grant to a known recipient with permitted-use, confidentiality and deletion terms. A company that understands the difference can license operational records while keeping its real secrets out of the package.
SourceX handles this inside the SourceX five-step transaction. In the Rights step, the supplier and its counsel identify trade secret material, such as pricing models, formulas or source code, and carve it out. In the Preparation step, personal and confidential details are removed from what remains. The SourceX Evidence Packet records provenance, licensing rights, permitted use, the privacy record and release authorization, and nothing is shared during the initial assessment.
Frequently asked questions
Does an enterprise AI plan fully protect our trade secrets?
No plan does that on its own. An enterprise plan with confidentiality and no-training terms removes the most obvious weakness, but courts still look at the company's overall measures. Access controls, a written policy, training and records of who used the tool for what remain part of the picture. Read the plan's actual terms, not its marketing page.
If an employee already pasted a secret into a consumer AI tool, is protection gone?
Not necessarily. The outcome may depend on what was disclosed, the tool's terms, how far the content could spread and how the company responded. Act quickly: document the disclosure, delete the content where the tool allows, preserve the terms in force at the time and ask counsel whether further steps, such as notice to a contract party, are needed.
Should we ban AI tools to be safe?
A blanket ban is hard to enforce and often pushes use onto personal phones and accounts, where the company has no visibility at all. Many companies get better protection by approving specific enterprise tools, restricting consumer ones for sensitive roles and training staff on what not to paste. Choose an approach counsel believes you can enforce and document.
Does this affect records we share with a buyer under a data license?
A data license is a different kind of disclosure. It goes to a known party under written confidentiality, permitted-use and deletion terms, which is the kind of measure trade secret law expects. It is still worth screening licensed packages for trade secret content and excluding anything the company wants to keep protected.
Do customer contracts add obligations here?
Often, yes. Many customer agreements and NDAs require you to protect the customer's confidential information and limit who receives it. Entering a customer's drawings, pricing or data into a consumer AI tool can raise a contract question separate from your own trade secrets. Review those agreements with counsel when you write the AI use policy.
Sources
- Under 18 U.S.C. 1839(3), information qualifies as a trade secret only if the owner has taken reasonable measures to keep it secret and it derives independent economic value from not being generally known. Source
- DOJ guidance states that trade secret protective measures need not be absolute but must be reasonable under the circumstances, citing examples such as limiting access on a need-to-know basis and requiring confidentiality agreements. Source
- The Defend Trade Secrets Act of 2016 created a federal civil cause of action for trade-secret misappropriation. Source
- An employer that fails to provide the 1833(b) whistleblower immunity notice may not recover exemplary damages or attorney fees against an employee who did not receive it; the requirement applies to contracts entered into or updated after the DTSA's enactment. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.