DeepSeek is being tested on the use of stolen data from AI companies

OpenAI and Microsoft are testing whether DeepSeek has been trained on stolen data from U.S. AI companies. Back in the fall of last year, it was revealed that individuals associated with DeepSeek were siphoning off huge databases of information using OpenAI API.

Microsoft as the largest investor of OpenAI stated about the possibility of this case. This may have violated OpenAI's terms of service, and it is also possible that the group was attempting to break data limits.

US Administration Reaction

The U.S. administration has floated the possibility that an AI company from China may have trained its DeepSeek R1 model based on OpenAI models. The learner model can achieve significant performance gains. No concrete facts and confirmations of such a situation have yet been announced to support the hypothesis.

In the OpenAI usage policy, there is a prohibition on copying the output data and further use of the information for the development of models. The OpenAI company has not yet reacted to such information. They only announced that they constantly take measures to protect intellectual property and work closely with the U.S. government to effectively respond to the illegal actions of competitors.

Initially, DeepSeek costs were advertised at around $5 million, but analysts have calculated that there is a strong possibility that the company only spent $500 million on the hardware part. In addition, there were still costs for final training, experiments, and previous versions of the project.

Now the U.S. government is investigating the possible risks of DeepSeek for the security of the country. In addition, users are advised to be quite careful with the use of DeepSeek.

According to DeepSeek's terms of use and privacy policy, the service stores and processes data related to IP, keystrokes, cookies, performance logs, operating systems, and more.