A company discovered during a routine security audit that engineers had been using ChatGPT to debug code and get development help.
They pasted proprietary algorithms, unreleased features, and internal system architecture into the AI. Dozens of engineers did this for months. Nobody thought it was a problem – they were just working more efficiently.
Then legal explained what happened. The company’s core competitive IP was disclosed to OpenAI. Trade secret protection may be lost because secrets stop being secret once disclosed. Competitors using the same AI could potentially access similar insights or reverse-engineer the approaches.
The company must now disclose to investors and the board that IP protection is compromised. And there’s no way to un-disclose information once it’s been fed to an AI system.
How Employees Use AI With Confidential Information
Employees across companies use AI tools to work faster without realizing they’re disclosing confidential information.
Engineers paste code into ChatGPT to debug errors or optimize performance. Product managers share unreleased roadmaps to draft strategy documents. Finance teams input proprietary financial models to build projections. Sales teams upload customer lists and pricing to generate proposals.
Each use feels productive. The AI helps solve problems faster. But every prompt containing confidential information is a potential trade secret disclosure that could destroy IP protection.
Most companies have no visibility into what employees are putting into AI tools because they’re using personal accounts on consumer platforms.
Why This Destroys Trade Secret Protection
Trade secrets are only protected if they remain secret. Once disclosed to third parties without confidentiality protections, trade secret status is lost.
When employees input proprietary information into AI platforms, they’re disclosing it to the AI vendor. Most AI platforms’ terms of service state they may use inputs to improve their models, which means the information could influence AI outputs for other users.
Even if specific code or data isn’t reproduced exactly, competitors using the same AI could receive suggestions influenced by your proprietary information. The AI learned patterns from your trade secrets and may reflect that learning in responses to others.
Courts have held that once trade secrets are disclosed without adequate protections, they lose legal protection permanently. You can’t reclaim trade secret status after disclosure.
Real Company Examples
Samsung discovered employees had leaked confidential source code and internal meeting notes into ChatGPT while using it for work tasks. The company banned ChatGPT immediately and had to assess what competitive damage occurred from the disclosure.
Other companies found during M&A due diligence that their development teams had been using AI with proprietary code for months. Buyers questioned whether the company’s IP had sufficient trade secret protection given the disclosures.
Law firms discovered attorneys had input confidential client information into AI tools to draft documents, potentially waiving attorney-client privilege and violating confidentiality obligations.
The common pattern is companies discovering the problem months after widespread employee use, when the damage is already done.
The AI Vendor Terms Problem
Most AI platforms’ terms of service explicitly state they may use customer inputs to train and improve their models.
OpenAI’s terms state they may use API inputs for training unless customers opt out. Consumer ChatGPT inputs are used for training by default. Anthropic, Google, and other providers have similar terms with various opt-out mechanisms.
Employees using consumer versions of AI tools with company information are disclosing trade secrets under terms that allow the AI vendor to use that information. Even enterprise versions with stricter terms still create disclosure to the vendor.
Most employees have never read these terms and don’t understand they’re potentially waiving trade secret protection with every prompt.
What Companies Should Do Immediately
Implement clear policies about AI use with confidential information. The default should be that employees cannot input trade secrets, proprietary code, customer data, or other confidential information into AI tools without approval.
For approved AI use, require enterprise versions with contractual protections about data use rather than consumer tools. Ensure contracts prohibit using company inputs for model training.
Train employees on what constitutes confidential information and why inputting it into AI destroys trade secret protection. Many employees don’t realize casual work help from AI creates legal issues.
Audit what’s already been disclosed. Review AI usage logs if available, interview employees about their AI use, and assess what proprietary information may have been compromised.
For critical IP, assume competitors using the same AI platforms may benefit from insights derived from your disclosures even if they don’t see your exact inputs.
The Investor Disclosure Problem
If significant trade secrets were disclosed to AI platforms, companies may have disclosure obligations to investors and boards.
Material changes to IP protection affect company valuation. If core competitive advantages are no longer protected as trade secrets due to AI disclosure, that’s material information investors need to know.
During fundraising or M&A, buyers will ask about AI use with proprietary information. If disclosure happened, it affects IP due diligence and company value.
Some companies have had to mark down valuations or delay fundraising while assessing the scope of trade secret compromise from employee AI use.
Ask Yourself This
Do employees use ChatGPT, Claude, or other AI tools for work? Have they input proprietary code, confidential data, or trade secrets into these tools? Does the company have policies preventing this?
If you can’t answer confidently, trade secrets have likely been disclosed and IP protection may be compromised.
Schedule a 30-minute AI Trade Secret Risk Assessment where we’ll review what confidential information may have been disclosed through AI use, show you what policies prevent future disclosures, help assess damage to trade secret protection, and create employee training on AI and confidentiality.
Because using AI to work faster feels productive until you discover employees disclosed your core IP to systems your competitors also use.
Brian Alexander
Founding Partner, VisionAI+ Consulting Group
Former General Counsel, AI Engineering Firm
Certified ISO 42001 Lead Auditor