The world of artificial intelligence is experiencing one of its most important battles to date. OpenAI and Microsoft are investigating DeepSeek, a Chinese startup that may have improperly used OpenAI’s models to train its own AI.
If confirmed, this would be a case of intellectual property misuse, trade secret appropriation, and potential breach of contractual terms. But what is really behind this dispute? Can a company claim ownership of knowledge generated by another AI? Let’s analyze it in detail.
DeepSeek and the technique of “model distillation”
To understand this conflict, we first need to talk about (model distillation). This technique involves training an AI using the responses generated by a more advanced model as a reference. Essentially:
- Thousands (or millions) of questions are asked to GPT-4.
- Its responses are collected.
- Then, a new model is trained by replicating that knowledge.
The problem is that if DeepSeek used GPT-4’s data without authorization, this could be an infringement of OpenAI’s intellectual property rights. But here arises the big question: can the responses of an AI be protected by copyright?
The potential legal infringements of DeepSeek
This case has three major legal fronts:
Copyright infringement?
In theory, the source code and the data used to train an AI can be protected by copyright. However, the responses generated by the AI are in a legal gray area. There is no clear consensus on whether they belong to the company that created the model or if they are in the public domain.
If OpenAI can prove that DeepSeek used GPT-4 responses on a massive and systematic scale to train its model, it could argue a copyright violation. However, this would be a precedent-setting case, as courts have not yet clearly defined whether AI-generated responses are protected by copyright.
Misuse of trade secrets
Another key front is the potential appropriation of trade secrets. If DeepSeek had access to confidential information from OpenAI (for example, internal leaks or reverse engineering of the model), this could violate the Defend Trade Secrets Act (DTSA) in the U.S.
The challenge here is proving it. OpenAI would need to demonstrate that DeepSeek illegally obtained internal information about its technology.
Violation of terms of service
If DeepSeek used OpenAI’s APIs for mass queries and to train its model, it would be violating OpenAI’s terms of service, which prohibit the use of its responses to train competing models. This would provide OpenAI with a clearer legal basis to file a lawsuit.
How can DeepSeek defend itself?
DeepSeek has not remained idle and has several strategies to defend itself:
DeepSeek has several strategies to defend itself against OpenAI’s accusations. First, the company could argue that its model was trained with its own data and from public sources, making it difficult for OpenAI to prove misuse of its technology. Additionally, DeepSeek could claim that its model is similar but not a copy, as the mere fact that two models generate similar responses does not necessarily mean one was copied from the other. Finally, another possible defense is that DeepSeek has released its model as open-source, which could serve as an argument to dismiss misappropriation claims. However, this strategy would not exempt it from liability if OpenAI manages to prove that its technology was used without permission.
How could the case end?
The OpenAI vs. DeepSeek case could be resolved in several ways. One of the most likely options is that OpenAI files an intellectual property infringement lawsuit, although the fact that DeepSeek is a Chinese company would complicate the process and the enforcement of any potential ruling. Another possibility is that the U.S. could impose government restrictions on DeepSeek, banning its operations in the country, as has happened with other Chinese tech companies like Huawei or TikTok. There is also the option of a settlement agreement, where both parties negotiate to avoid a prolonged and costly lawsuit. Finally, this case could drive changes in AI regulation, setting a legal precedent on intellectual property in artificial intelligence and accelerating the development of specific regulations for model training
The OpenAI vs. DeepSeek case is not just a conflict between two companies, but a clash of philosophies: privacy and control vs. open-source and free access. The resolution of this case could change the game in AI, defining who owns artificial intelligence and how its knowledge can be used.
Whatever the outcome, it is clear that we are on the brink of a new era in AI regulation. And this is just the beginning.
