Trade secrets are only protected if the business has taken reasonable steps to keep them secret. One employee pasting a confidential document into ChatGPT does not normally remove that protection. But if the business lets employees use AI tools freely, without rules and without an enterprise agreement, it becomes hard to show reasonable steps when the secret later has to be enforced.
A sales director pastes the customer list and next year’s pricing strategy into ChatGPT to get help with a presentation. A developer asks a coding assistant to improve the company’s most valuable algorithm. Both happen every day in Norwegian businesses. The question is whether the business thereby risks losing legal protection for the information.
The answer is that a single isolated incident is rarely decisive. The real risk is that the business lacks control, and therefore cannot show that it has protected its secrets when it one day matters.
When is information a trade secret under the law?
Information is a trade secret when three conditions are met at the same time. This follows from Trade Secrets Act § 2 of the Norwegian Trade Secrets Act (forretningshemmelighetsloven), which implements the EU Trade Secrets Directive in Norwegian law. The Act has applied since 1 January 2021.
| Condition | What it means in practice |
|---|---|
| Secret | The information is not generally known or readily accessible, either as a body or in the precise configuration of its components |
| Commercial value | Its value derives precisely from the fact that it is secret |
| Reasonable steps | The holder has taken reasonable steps to keep it secret |
Typical examples are source code, customer lists, pricing models, bid strategies, product recipes, business plans and technical solutions. Under Trade Secrets Act § 2(2) the general experience and skills an employee has acquired in the course of employment are not trade secrets.
The conditions are cumulative. If one of them is missing, the business does not enjoy the protection of the Act, however valuable the information is.
What do “reasonable steps” mean when employees use AI tools?
Reasonable steps means that the business has actively done something to protect the information, and can document it. The law does not require complete security. What is reasonable depends on a specific assessment of how valuable the information is, how large the business is and what threats it faces.
Common measures are confidentiality agreements, access control, marking of confidential documents, internal guidelines and training. AI tools are now such an obvious channel for leaks that most businesses have to address them. A business that has neither rules for the use of AI nor knowledge of which tools its employees use will be in a weak position when it tries to convince a court that it has protected a particular piece of information.
The law does not protect what the business itself has left open. Reasonable steps are the price of protection.
There is currently no Norwegian case law on how the use of AI tools affects the assessment of reasonable steps. The assessment above is based on the wording of the Act and the general understanding of the condition.
Do you lose protection for trade secrets when someone uses ChatGPT?
Not automatically. The fact that one employee has put a confidential document into a chatbot does not normally make the information generally known. The provider is one party, not the public at large, and protection does not lapse because a single person breaks the business’s rules.
The risk nonetheless increases in three ways. First, systematic and uncontrolled use may count against the business having taken reasonable steps. Second, information used to train a language model could in principle be reproduced for other users later. How great this risk is in practice, and whether it could make the information readily accessible within the meaning of the Act, has not been settled. Third, the business loses control over where the information is stored, who has access to it and how long it is kept.
It is the first point in particular that affects the bottom line. Without protection under the Act, the business cannot seek an injunction, damages or criminal penalties when a former employee or a competitor walks off with the information.
Why does the choice of licence matter so much?
The choice between a free or consumer version and an enterprise agreement is often the single most important measure. The difference lies in what the provider may do with the content.
| Consumer version | Enterprise agreement or API | |
|---|---|---|
| Training on content | Often yes, unless the user opts out | By default no, with the major providers |
| Confidentiality | Standard terms for private individuals | Contractual duty of confidentiality |
| Data processing agreement | Not normally available | Available |
| Administration | The employee is in control | The business controls users and settings |
| Documentation | Little or none | Agreement and settings can be produced |
OpenAI states, for example, that conversations in the consumer versions of ChatGPT may be used to improve the models unless the user opts out, whereas content from the enterprise versions and the API is not used for training by default. The terms change regularly and vary between providers, so they must be checked for each tool before it is rolled out.
An enterprise agreement is not just a security measure. It is also documentation. When the business later has to show that it has taken reasonable steps, a signed agreement with a duty of confidentiality and an opt-out from training is far stronger evidence than a verbal reminder to “be careful”.
Can using AI tools breach non-disclosure agreements?
Yes, and this is often the most concrete risk. Many businesses have received information from customers, suppliers and partners under a non-disclosure agreement (NDA). Such agreements usually prohibit making the information available to third parties without consent. Putting the information into an external AI tool may be exactly that.
A breach may give rise to liability in damages, trigger an agreed liquidated damages clause and harm the customer relationship. This applies regardless of whether the information meets the conditions of the Trade Secrets Act. Review your most important NDAs and establish whether they allow the use of subcontractors and cloud services.
The same risk arises when management uses AI to draft contracts and pastes in confidential negotiating positions. Read more in the article on vibe lawyering and the risks of AI-drafted contracts.
What can the business do when secrets are misused?
If the information qualifies as trade secrets, the Act provides strong remedies. This is where the reasonable steps pay off.
- Injunction. A person who has infringed may be prohibited by judgment from repeating the act under
Trade Secrets Act § 5. Legal action must be taken no later than three years after the holder obtained, or should have obtained, the necessary knowledge of the infringement and the infringer. - Compensation and damages. For intentional or negligent infringement, the holder may under
Trade Secrets Act § 8claim whichever is most favourable of a reasonable licence fee plus damages, damages for the loss suffered, or compensation corresponding to the infringer’s gain. - Criminal penalties. Infringement is punishable by fines or imprisonment of up to one year under
Trade Secrets Act § 9, and aggravated infringement by up to three years underTrade Secrets Act § 10.
Infringement includes, among other things, unlawful use or disclosure of information that someone has obtained in an employment or business relationship, see Trade Secrets Act § 3(2). For employees, the decisive question is whether the use was unlawful. An employee who has followed the business’s own rules, or who has never been given any, will struggle to have acted unlawfully.
At the same time, guidelines on AI use must not be worded so as to restrict employees’ right to blow the whistle on wrongdoing under Working Environment Act § 2 A-1 of the Working Environment Act (arbeidsmiljøloven).
What should the business do?
- Map what is secret. Identify the information that actually has commercial value, and mark it. What has not been identified is hard to protect and hard to prove.
- Choose approved tools with an enterprise agreement. Secure a contractual duty of confidentiality, an opt-out from training, a data processing agreement and administrative control.
- Adopt internal rules for AI use. The rules should state which tools are approved and what information must never be entered. See the article on AI policies for employees.
- Review your NDAs. Check whether agreements with customers and partners permit the use of AI tools and cloud services, and update your own templates.
- Provide training. Rules nobody knows about are not a reasonable step.
- Document the measures. Keep agreements, guidelines, training logs and access controls. It is the documentation that decides the case when protection has to be enforced.
The question of who owns what the AI tool produces is a different one from protecting what goes into it. It is covered in the article on copyright in AI-generated content. More articles on the subject are available on the topic page on artificial intelligence.
Questions and answers
Does ChatGPT use what we type in to train the model?
It depends on which version you use. According to OpenAI, the models are trained on conversations in the consumer versions unless the user opts out, whereas content from the enterprise versions and the API is not used for training by default. The terms change regularly and must be checked when contracting.
Is it enough to ban the use of AI tools in the business?
A ban is a measure, but it only works if it is complied with. Employees without an approved alternative often turn to private accounts. An approved tool with an enterprise agreement, clear rules and training will usually give better protection than a ban nobody follows.
Can we claim damages from an employee who has put secrets into an AI tool?
Only if the infringement is unlawful and intentional or negligent. An employee who has followed the business's own rules, or who has never been given any rules, will not normally have acted unlawfully. Clear guidelines are therefore also a precondition for being able to respond.
Next legal review: 15 January 2027