OpenAI accuses DeepSeek of distilling US models to train its AI, warns lawmakers of security risks

OpenAI Accuses DeepSeek of Distilling US Models to Gain an Edge

The practice of distillation poses a business threat to American companies and raises national security risks, including the potential for misuse of AI models in high-risk areas. OpenAI has warned US lawmakers that its Chinese rival DeepSeek is using unfair and increasingly sophisticated methods to extract results from leading US AI models to train the next generation of its breakthrough R1 chatbot, according to a memo reviewed by Bloomberg News. In the memo, sent Thursday to the House Select Committee on China, OpenAI said that DeepSeek had used so-called distillation techniques as part of "ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs." The company said it had detected "new, obfuscated methods" designed to evade OpenAI's defenses against misuse of its models' output. OpenAI began privately raising concerns about the practice shortly after the R1 model's release last year, when it opened a probe with partner Microsoft Corp. into whether DeepSeek had obtained its data in an unauthorized manner, Bloomberg previously reported. In distillation, one AI model relies on the output of another for training purposes to develop similar capabilities. Distillation, largely tied to China and occasionally Russia, has persisted and become more sophisticated despite attempts to crack down on users who violate OpenAI's terms of service, the company said in its memo, citing activity it has observed on its platform. Since DeepSeek and many other Chinese models don't carry a monthly subscription cost, the prevalence of distillation could pose a business threat to American companies such as OpenAI and Anthropic PBC that have invested billions of dollars in AI infrastructure and charge a fee for their premium services. That imbalance risks eroding the US advantage over China in artificial intelligence. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg may send me offers and promotions. Plus Signed UpPlus Sign UpPlus Sign Up By submitting my information, I agree to the Privacy Policy and Terms of Service. OpenAI also highlighted other national security risks raised by DeepSeek's gains, including that its chatbot had censored results about topics considered controversial by the Chinese government such as Taiwan and Tiananmen Square. When capabilities are copied through distillation, OpenAI said, safeguards often fall to the wayside, enabling more widespread misuse of AI models in high-risk areas like biology or chemistry. Representative John Moolenaar, the Republican chair of the House China committee, said in a statement Thursday "this is part of the CCP's playbook: steal, copy, and kill," referring to the Chinese Communist Party. "Chinese companies will continue to distill and exploit American AI models to their advantage, just like when they ripped off OpenAI to build DeepSeek." OpenAI declined to comment on the memo. Spokespeople for the Chinese embassy in Washington and for DeepSeek didn't immediately respond to requests for comment. Read more Anthropic Clamps Down on AI Services for Chinese-Owned Firms OpenAI's memo to the House China committee suggests that its efforts to block distillation have failed to eliminate the problem. The company said an internal review suggests that accounts associated with DeepSeek employees sought to circumvent existing guardrails by accessing models through third-party routers to mask their source. DeepSeek employees have also developed code to access US AI models and obtain outputs in "programmatic ways," OpenAI said. It also points to networks of "unauthorized resellers of OpenAI's services," also designed to evade the company's controls. White House AI Czar David Sacks has previously warned about Chinese distillation tactics, telling Fox News last year that DeepSeek was "squeezing more juice" out of older chips while also citing "substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI's models." OpenAI's warning about distillation also comes as many in Washington remain concerned that access to advanced AI chips may also accelerate DeepSeek's progress. At the end of last year, President Donald Trump moved to ease chip restraints and allow Nvidia to sell its H200 processors, chips that are about 18 months behind the leading Blackwell versions. Since the R1 release, DeepSeek has only put out minor upgrades, even as competitors in the US and China have launched a flurry of new models. Bloomberg has reported that DeepSeek is developing an agent-based model to compete with OpenAI and other rivals and had hoped to release it in late 2025. Shortly after the R1 release, US authorities opened a probe into whether DeepSeek circumvented US export controls by purchasing chips via Singapore. Records recently obtained by the House China committee show that Nvidia Corp. provided technical support to help DeepSeek improve and co-design its R1 model. The DeepSeek-V3 base model required only 2.8 million H800 GPU hours for its full training. Those processors were allowed to be sold to China for a few months in 2023, until a rule later that year halted sales. "DeepSeek should have been a wake-up call about the dangers of selling advanced semiconductor chips to the CCP. Using less powerful Nvidia chips, China developed the most advanced open-source models on the planet," Republican Representative Michael McCaul, who formerly led a House panel with export control oversight, said last year in the wake of the H200 decision. "I shudder to think of what they might do with more advanced hardware like the H200 chips."

OpenAI says China's DeepSeek trained its AI by distilling US models, memo shows - The Economic Times

OpenAI has warned US lawmakers that Chinese artificial intelligence startup DeepSeek is targeting the ChatGPT maker and the nation's leading AI companies to replicate models and use them for ⁠its own ⁠training, a memo seen by Reuters showed. Sam Altman-led OpenAI accused DeepSeek of "ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs." The technique, known as distillation, involves having an older, more established and powerful AI model evaluate the quality of the answers coming out of a newer model, effectively transferring the older model's learnings. In the memo ⁠sent to the US House Select Committee on Strategic Competition between the US and the Chinese Communist Party on Thursday, OpenAI said: "We ⁠have observed accounts associated with DeepSeek employees developing methods to circumvent OpenAI's access restrictions and access models through obfuscated third-party routers and other ways that mask their source." "We also know that DeepSeek employees developed code to access U.S. AI models and obtain outputs for distillation in programmatic ways," the memo added. DeepSeek and its parent company High-Flyer did not immediately respond to Reuters' requests for comment. Hangzhou-based DeepSeek shook markets early last year with a set of AI models that rivaled some of the best offerings from the US, fuelling concerns in Washington that ⁠China could catch up in the AI race despite restrictions. OpenAI said that Chinese large language models are "actively cutting corners when it comes to safely training and deploying new models." Silicon Valley executives have previously praised models named DeepSeek-V3 and DeepSeek-R1, which are available globally. OpenAI said it proactively removes users who appear to be attempting to distill its models to develop rival's models.

OpenAI calls out DeepSeek for copying its model for self training: Here's what's happening

OpenAI said it had identified accounts allegedly linked to DeepSeek staff trying to get around access limits. OpenAI has raised concerns with US lawmakers, alleging that Chinese AI startup DeepSeek has been attempting to replicate the work of the nation's leading AI companies. According to a memo seen by Reuters, the company claims these actions are aimed at copying advanced AI capabilities and using them to train competing systems. The memo was sent to the US House Select Committee on Strategic Competition between the US and the Chinese Communist Party, where OpenAI outlined what it described as attempts to bypass safeguards designed to protect its models. The company said it had identified accounts allegedly linked to DeepSeek staff trying to get around access limits. 'We have observed accounts associated with DeepSeek employees developing methods to circumvent OpenAI's access restrictions and access models through obfuscated third-party routers and other ways that mask their source,' the memo stated. 'We also know that DeepSeek employees developed code to access U.S. AI models and obtain outputs for distillation in programmatic ways' Also read: OpenAI unveils ultra-fast GPT 5.3 Codex Spark model for real-time coding At the centre of the dispute is a technique called distillation. This method allows developers to use outputs from an old, more established and stronger AI system to train a newer model. OpenAI alleges that this approach is being used to 'free-ride on the capabilities developed by OpenAI and other US frontier labs.' The company also stated in the memo that Chinese large language models are 'actively cutting corners when it comes to safely training and deploying new models.' Also read: Apple iPhone 17e vs Google Pixel 10a: Price, camera, display, battery and other leaks compared Deepseek drew major attention last year after releasing AI models that appeared to rival some of the most advanced US offerings. Their performance raised concerns in Washington that China could catch up in the AI race despite restrictions, as per the report.