Microsoft suspected Deepseek in the training of AI on data stolen from Openai

Microsoft and Openai will conduct an investigation on whether the organization related to the Chinese laboratory of the DEEPSEEK artificial intelligence laboratory has done the theft of data from Openai. This was reported by Bloomberg with reference to its own sources.

Image source: deepseek.com

In the fall, experts of the security department of Microsoft noticed that some persons who, according to the corporation, may be associated with Deepseek, request large amounts of data through the Openai API, Bloomberg sources said. Openai sells access to the API – this tool allows third -party developers to integrate the Openai artificial intelligence models into their applications. Microsoft as a technological partner and the largest Openai investor notified the company about what was happening. This activity may violate the conditions of maintenance of Openai or indicate that the company related to the Chinese laboratory tried to get around the restrictions on the amount of data that Openai customers can receive.

Previously, Deepseek introduced the open model of AI R1, imitating the course of a person’s thoughts. The project stirred up the market, which is dominated by Openai and other American companies, including Google and Meta✴. According to the creators, the model can compete or exceeds the projects of leading American developers, and its training cost radically cheaper. As a result, there was a threat of American dominance in the II industry, and on Monday, January 27, the market capitalization of US technological companies, including Microsoft, Nvidia, Oracle and Alphabet, sank almost $ 1 trillion.

On the eve of January 28, the adviser to US President Donald Trump (Donald Trump) on AI David Sacks said that there is “significant evidence” of the use of OpenAi models in DeepSeek technologies. In an interview, Mr. Sax spoke about the distillation method, allowing one AI model to study on the data of another for the development of similar capabilities. “We know that companies from the PRC – and others are constantly trying to distilly models of leading American companies in the field of AI. As a leading developer of AI, we accept countermeasures to protect our intellectual property, including a thorough process of deploying advanced capabilities of produced models, and are sure that as moved forward, it is critically important to cooperate with the US government in order to most effectively protect the most functional models from the attempts of the enemy And competitors to take possession of American technologies, ”commented on the Saxi statement in Openai.