DeepSeek suspected of stealing OpenAI data

Appointed by Donald Trump as the ‘czar’ for AI and cryptocurrencies, David Sacks said there is “substantial evidence” that the startup DeepSeek used data from OpenAI’s models. He likened it to theft.

AI & Crypto Czar @DavidSacks says there’s “substantial evidence” that DeepSeek distilled knowledge out of OpenAI’s model. pic.twitter.com/e3ppKa2UWJ

— CAPITAL (@capitalnewshq) January 28, 2025

According to him, the Chinese company used outputs from OpenAI’s models to train its own.

“I don’t think OpenAI is very happy about this,” Sacks said.

Microsoft and OpenAI have begun investigating the matter, according to the media.

Microsoft security experts in autumn 2024 observed people linked to DeepSeek “siphoning large amounts of data” via OpenAI’s API.

DeepSeek R1’s official developer documentation uses the OpenAI package.

DeepSeek’s surge

Over the weekend of 25–26 January, the DeepSeek-R1 model sparked a frenzy and triggered sell-offs across equity and crypto markets. The company’s chatbot surged up the rankings of free apps in the US App Store, overtaking ChatGPT.

The appeal lies in offering top-tier performance at far lower cost than American models. OpenAI CEO Sam Altman responded to the new rival, calling the much-discussed DeepSeek-R1 “impressive” while pledging to surpass the Chinese competitor.

“Obviously, we will deliver much better models. Besides, it’s great to have a new competitor! We will pull forward several releases,” he added.

In a comment to ForkLog, cyber.fund principal Stepan Gershuni said DeepSeek could shake OpenAI’s dominance in artificial intelligence.

“This has already happened. […] But AI leaders change roughly every month. Google will release a model, then Anthropic, then OpenAI, then LLaMA [Meta], and now DeepSeek. Their model is not the best today. It is unlikely to remain the leader. I am sure we will see far more competition, far more other participants. But nothing terrible has happened to the American competitors yet,” he noted.

Gershuni also agreed that OpenAI could deliver a worthy response to DeepSeek, as Altman promised.

“Yes, of course. We will see this more than once. Moreover, both companies will pull attention to themselves many more times. It’s only just beginning,” the expert said.

In his view, intensifying competition will produce a better product for the market.

“[…] In AI this also has a certain political tinge, a kind of global domination. Whether it will be China, the US or a decentralised internet is unknown […],” he believes.

Meanwhile, the US Navy has already barred personnel from using DeepSeek’s AI technology, CNBC reported, citing “potential security and ethical issues”.

For now, OpenAI and DeepSeek each have strengths and weaknesses, according to Gershuni. Among ChatGPT’s advantages, he cited a more polished product that includes:

• a better app;
• an advanced voice mode;
• the AI agent ‘Operator’;
• the online GPT Store.

Among the drawbacks are a closed, paid system.

DeepSeek, by contrast, is free and open-source. The model can run locally on a device without the internet.

Market impact

DeepSeek’s emergence raised questions about the wisdom of multibillion-dollar investments in GPUs, as the model offered performance comparable to top-tier systems without using large numbers of expensive chips. Nvidia shares fell 17% during the 27 January session.

Online, some drew parallels with Cisco, whose shares slumped after a sharp run-up during the dot-com bust.

Gershuni believes Nvidia is not headed for a Cisco-like fate.

“[…] There are no reasons today for Nvidia’s shares to crash severely. Maybe it will be worth several hundred billion less, but that changes nothing. Nvidia is still the absolute leader, effectively a monopolist, and will likely remain so for some time. I would sooner buy Nvidia shares, but this is, of course, not financial advice,” he stressed.

After the double-digit drop, the chipmaker’s shares rebounded 9% during the 28 January session.

Niles Investment Management’s Dan Niles believes the correction in the stock is not over.

“I think investors should be very careful believing this is the bottom,” he stressed.

If revenue-growth estimates fall from 50% to 20–30%, there is significant downside potential, the portfolio manager said. Earlier, Nvidia sought to rebut suggestions of a potential drop in demand for GPUs.

“Inferences require a significant quantity of our graphics processors and high-performance networks,” company representatives said.

Analysts have called the frenzy around DeepSeek premature.

Earlier, the startup unveiled a set of new multimodal AI models that, it claims, outperform OpenAI’s DALL·E 3.