trade secret
A trade secret is non-public corporate knowledge or data that is valuable precisely because of its secrecy, and which the owner protects with reasonable steps.
In plain terms
Think of your grandmother's treasured pastry recipe: it is written down nowhere, yet it is hers until she gives it out. A trade secret is the same on a corporate scale: a recipe, customer list, price calculation, manufacturing process or source code that is worth money because competitors do not know it.
Unlike a patent, it does not require registration, has no official fee, and no fixed expiration date. In return, it protects only as long as the secret remains truly secret and the company takes measures: passwords, non-disclosure agreements, and restricted access. These measures can, of course, cost money.
In the AI world, this is important from two angles. Many AI companies treat their model weights, internal settings, training data details, and system instructions as trade secrets. However, there are also open-weight models whose weights are intentionally published. The other angle concerns you directly: if you paste a corporate document into a public chatbot or grant an agent access to corporate folders, the secret may escape the company's control.
How it works
- Three conditions must be met simultaneously: the information is not common knowledge and not easily accessible, it has economic value due to its secrecy, and the holder takes reasonable steps to preserve it.
- Protection is automatic: there is no office, no registry, and no fee. In case of dispute, the company must prove that the data was truly treated as a secret.
- It is illegal if someone acquires, uses, or discloses something unfairly, for example through theft, deception, or breach of a confidentiality agreement.
- It is not infringement if someone independently arrives at the same result, or reverse-engineers how a legally acquired product works, provided a contract does not prohibit this.
- In case of infringement, the court may ban use, order the withdrawal or destruction of infringing products, and award damages.
An everyday example
Before the deadline, you paste the company's draft price quote into a free chatbot to have it worded more nicely. The text then lands on an external provider's server and, depending on the settings, may even be used to train the model. This is why many workplaces require that confidential material only be entered into company-approved AI tools.
Who is affected
A common misconception
Limits and risks
- A leaked secret cannot be undone; compensation rarely replaces lost competitive advantage.
- If the company did not take reasonable steps to protect the data (no policy, no access restrictions), the court may not recognize the data as a trade secret.
- A trade secret is not an unlimited shield: authorities and courts cannot be refused information required by law by invoking it, and the law also protects whistleblowers acting in the public interest.
- An AI agent can access as many secrets as the permissions you grant it, so always open only the folders and systems necessary for the task.
Similar, but not the same
- szabadalom – To obtain a patent, the invention must be disclosed publicly; in return, it grants exclusive rights for up to 20 years from the filing date, whereas a trade secret remains hidden and is protected as long as it stays secret.
- personal data – Personal data protection (GDPR) protects the individual, while trade secrets protect a company's competitive advantage. A customer list containing individuals can be both.
- copyright – Copyright protects the formulation and form of a work, whether public or not. It does not protect the idea or the information itself. Trade secrets, however, protect the information itself, but only as long as it remains secret.
Related concepts
szellemi tulajdon, data security
know-how, system prompt, model weights, training data
szabadalom, copyright, non-disclosure agreement, transparency, EU MI-rendelet
A brief history
Internationally, the 1994 TRIPS Agreement mandated the protection of "undisclosed information". The European Union harmonized the rules in 2016 with Directive 2016/943, which Hungary transposed with the 2018 Act LIV on the protection of trade secrets.
Where things stand
A long-established legal institution that has returned to the spotlight due to AI. By the end of 2026, the main debate concerns how much AI developers must disclose about their training data and models under the EU AI Act, and what they can keep as trade secrets.
In the news
- Hungarian professional publications: shadow AI, deepfakes and data protection risks at companiesNov 3, 2026 · Jog
- US federal appeals court: law firm's AI use does not justify cutting legal feesNov 3, 2026 · Jog
- Apple and OpenAI trade secrets lawsuit formally under way, trial set for July 2027Oct 29, 2026 · Jog
- Judge grants Apple limited expedited discovery in OpenAI trade secrets caseOct 13, 2026 · Jog
- AI drug discovery tools carry risky contract terms as USPTO inventorship rules changeSep 28, 2026 · Health
- Apple sues OpenAI over trade secrets; OpenAI hits back publicly and seeks dismissalAug 4, 2026 · Jog
Also known as
trade secret · trade secret · trade secrets · kereskedelmi titok · confidential business information · confidential business information
Sources
- Act LIV of 2018 on the Protection of Business Secrets (National Legislation Database)
- Trade secret – Wikipedia
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.