Can Manufacturers Use AI Without Exposing Trade Secrets?

Manufacturers can use AI without losing trade secret protection, but it takes more than picking a popular tool. Feeding proprietary formulas, CAD files, or process data into a public AI chatbot can strip away the legal protection those secrets rely on, because U.S. trade secret law only protects information the owner actively keeps secret. The fix isn't avoiding AI. It's choosing tools and contracts that keep your data inside your own walls.

What happens to a trade secret once it's typed into a public AI tool?

Once an employee pastes a formula, drawing, or process step into a consumer AI chatbot, the company usually loses meaningful control over that information. Research from Jones Day on generative AI and trade secrets found that many AI providers can retain, review, or use submitted inputs to train their models, and that users often can't retrieve or delete what they've already submitted. If the provider suffers a breach, or if the end-user license lets the provider "review, release, or sell" information, a trade secret can end up disclosed to people who were never supposed to see it.

That matters because trade secret status isn't automatic. The Defend Trade Secrets Act (DTSA), the federal law that lets companies sue over trade secret theft, only protects information the owner has taken "reasonable measures" to keep secret. State trade secret statutes modeled on the Uniform Trade Secrets Act (UTSA) use nearly identical language. Courts decide what counts as "reasonable" case by case, and pasting a formula into a tool with no contractual data protections is a hard case to win.

Which manufacturing trade secrets are actually at risk?

The information manufacturers feed into AI tools is rarely abstract. It's usually one of a short list: product formulas and recipes, CAD and engineering drawings, process parameters and yield data, supplier pricing, or quality-control thresholds that took years to tune. A 2026 analysis from Foley & Lardner on IP in smart manufacturing points out that sensor outputs, digital twin telemetry, and AI inference logs are becoming strategic assets in their own right — and frequent sources of dispute when a shared production line or a third-party AI vendor is involved. An engineer asking an AI tool to "optimize this mixing ratio" has just handed over exactly the kind of detail a competitor would pay to see.

What should a manufacturer's AI vendor contract actually require?

A vendor contract is where most of the real protection gets decided, long before anyone opens the AI tool. Based on guidance from Foley & Lardner and Jones Day, a manufacturer's AI agreements should include:

• A contractual bar on using submitted data to train the vendor's models, not just a policy statement
• Confidentiality terms that extend to the vendor's subcontractors and cloud infrastructure
• Audit rights, so the manufacturer can verify how its data is actually handled
• Data deletion or return obligations that trigger automatically at contract termination
• Clear ownership language for any new data the AI tool generates, like inference logs or derived process recommendations

Enterprise agreements can sometimes negotiate these terms with major AI vendors. The simpler path for a lot of manufacturers, though, is choosing a tool where the question doesn't come up in the first place, because the data never leaves the building.

Does on-premise or private AI solve the trade secret problem?

It solves the part of the problem that's actually solvable with a product choice. When an AI system runs on hardware inside a manufacturer's own facility, on its own network, there's no vendor server to retain inputs, no third-party training pipeline, and no breach at a cloud provider that exposes someone else's data alongside yours. That's the model ForgeAI uses in manufacturing settings: the appliance sits on the customer's network, works only from documents and process files the company has approved, and never sends that content to an outside server for processing or model training. You can see how that fits specific industries, including manufacturing, on ForgeAI's industries page (opens in a new tab).

On-premise deployment doesn't replace a trade secret program by itself. It removes one specific failure mode: the uncontrolled third-party disclosure that generic cloud AI tools can create. The legal "reasonable measures" test still expects the rest of a trade secret program to be in place around it.

What else should manufacturers do besides choosing the right AI tool?

Tool choice is one layer. The natlawreview analysis of trade secret risk in smart manufacturing recommends pairing it with a few habits most manufacturers already know from physical security, applied to AI:

1. Restrict AI access by role, so the floor-level operator and the process engineer don't have the same visibility into formulas or yield data.
2. Require NDAs with AI vendors, contractors, and any third party who touches the production data pipeline, with explicit return-or-destroy terms at the end of the relationship.
3. Run periodic trade secret audits to confirm which systems actually hold sensitive process data, not just which systems were supposed to.
4. Train staff specifically on what counts as a trade secret in your plant, since "don't share confidential info" means little without concrete examples like formulas, supplier terms, and yield tuning.

None of this is about banning AI. A flat ban tends to push usage underground, where the company has even less visibility into what's being typed into which tool. The goal is matching access to sensitivity, the same way a manufacturer already controls who walks onto the production floor.

Manufacturers don't have to choose between using AI and protecting what makes their products theirs. The companies getting this right are the ones treating AI tool selection as a trade secret decision, not just an IT purchase, and asking the same "where does this data actually go" question they'd ask about any new vendor on the production floor.

Frequently asked questions

Does using AI automatically void trade secret protection? No. Trade secret protection depends on whether the owner took reasonable measures to keep the information secret, not on whether AI was involved at all. The risk comes from how a specific tool handles submitted data — whether it retains, trains on, or could expose that input to others — not from AI use in general.

Is a company's internal "AI usage policy" enough to protect trade secrets? A written policy helps, but on its own it doesn't control what a vendor's AI tool does with data once it's submitted. Courts weighing "reasonable measures" look at actual technical and contractual controls, not just internal guidance employees may or may not follow.

Can a small or midsize manufacturer actually afford on-premise AI? Pricing varies by vendor and scope, so it's worth getting a specific quote rather than assuming the cost. The more useful comparison is against the potential cost of a trade secret dispute or a competitor gaining access to proprietary process data, which is typically the bigger number.

What's the single biggest AI-related trade secret mistake manufacturers make? Letting individual employees decide, tool by tool, what's safe to paste into a public AI chatbot. Without a specific, enforced policy and a vendor contract that addresses data handling, that decision gets made informally, dozens of times a day, by people who aren't thinking about DTSA case law.