Trade Secret
A trade secret is confidential business information that gives a company an economic advantage as long as it remains secret. Unlike a patent, a trade secret never has to be disclosed publicly and is theoretically protected for an unlimited period of time.
A trade secret is confidential information that generates money or market advantages for a company. This can be a recipe, a production method, or an algorithm. What matters is: the information is not publicly known, the company takes measures to keep it secret, and it has a measurable economic value. If all three conditions are met, it is legally recognized as a protectable trade secret in most countries. Anyone who steals it or discloses it without authorization is committing an offense.
Trade secrets as a corporate weapon
A patent protects an invention, but it comes at a price: the company must publicly describe the invention. Once the protection period expires – usually 20 years – anyone may use it. A trade secret works the other way around. Anyone who publishes nothing also loses no rights through the passage of time. The Coca-Cola recipe is said to have been secret for over 130 years – a patent would have expired long ago.
For technology companies, trade secrets are often more valuable than patents. An algorithm can be patented, but then competitors know it. If it remains secret, they cannot legally copy it – and don’t even know exactly how it works. This is precisely why AI companies like OpenAI or Google guard the weights of their models – that is, the billions of numerical values that make up a trained model – as trade secrets.
How protection works in practice
The state does not automatically protect a trade secret the way it protects a registered patent. The company itself must take action: employees sign confidentiality agreements, server rooms have access controls, documents are classified as confidential. Only those who can demonstrably prove they took protective measures can claim in court that a secret was stolen from them.
In Germany, the Trade Secrets Act (GeschGehG) has regulated this area since 2019, while in the US it is the Defend Trade Secrets Act. Both laws require the company to be able to demonstrate 'appropriate measures' for protection. A mistake here can lead a court to revoke the protected status – even if the information is still not publicly known.
A common misconception: anyone who independently discovers a trade secret through their own research is allowed to use it. Protection only applies against theft, betrayal, and industrial espionage – not against parallel development. This fundamentally distinguishes trade secrets from patents, which also bind independent inventors.
Trade secrets in AI news
In the AI industry, the term regularly comes up in legal proceedings. When former Google engineer Anthony Levandowski was convicted in 2019, the case involved stolen design plans for self-driving cars – trade secrets in the technical sense. In 2023, Google sued an employee who allegedly passed on training data and model architectures to a Chinese company.
The debate around open source also touches on this topic. When a company publishes the source code of its model, it gives up part of its trade secrets – voluntarily, in order to gain trust or developer communities. Meta does this with its Llama model family. OpenAI, on the other hand, does not publish the code for GPT-4. Both decisions are strategic, not a technical necessity.
This is relevant for investors and analysts: the value of an AI company often depends on how well it protects its secrets – and whether it can prove, in the event of a dispute, that it has done so.