
Image: © kritsapong_j/Stock.adobe.com
OpenAI and Microsoft are investigating whether DeepSeek obtained the ChatGPT-maker’s tech without authorisation.
A day after OpenAI CEO Sam Altman praised DeepSeek for its “impressive” and cheap R1 reasoning model, the company said that it found evidence that the Chinese start-up used OpenAI’s models to train its own open-source model, according to a Financial Times report.
DeepSeek’s R1 has made waves in the technology sector over the past week, showcasing capabilities on a par with industry heavyweights such as OpenAI’s ChatGPT and Anthropic’s Claude 3.5 Sonnet while needing only $5.6m to train the model – a fraction of what it costs its US competitors.
Its release even elicited a response from US president Donald Trump, who called that the start-up’s advancements a “positive”, but said that this should be a “wake-up call” for US industries.
However, the Financial Times now reports that OpenAI has some evidence that DeepSeek ‘distilled’ or leveraged outputs from a large model, here, ChatGPT, into its smaller model, reducing costs and latency.
Meanwhile, Bloomberg reported that the ChatGPT-maker and Microsoft – OpenAI’s largest investor – are jointly investigating whether DeepSeek obtained OpenAI’s technology in an unauthorised manner.
Sources told the publication that last year Microsoft’s security researchers observed persons they believed to be connected to DeepSeek taking a large amount of data using OpenAI’s application programming interface (API), a service that developers need to pay to use.
However, OpenAI itself has come under fire numerous times for allegedly training its models using third-party data without consent. Currently, OpenAI and Microsoft are facing a lawsuit by the New York Times for the alleged copyright infringement of millions of the publication’s articles.
While, in a now dismissed lawsuit, US law firm Clarkson alleged that OpenAI ignored any legal means of obtaining training data for its chatbot, choosing rather to gather it from the web without consent from the millions who have uploaded content.
DeepSeek has also caused uproar in Washington, with the White House press secretary Karoline Leavitt telling reporters yesterday (28 January) that the National Security Council is reviewing DeepSeek’s security implications.
She said that “Trump believes in restoring AI dominance”, adding that the president has taken “very strong” executive actions to reduce regulations on the US AI industry.
Moreover, CNBC reported yesterday that the US navy has instructed its members to avoid using DeepSeek.
In a email issued to “shipmates” last week, the Navy said that DeepSeek’s AI was not to be used “in any capacity” due to “potential security and ethical concerns associated with the model’s origin and usage”.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.