OpenAI Accuses Chinese AI Firm DeepSeek of Unauthorized Model Distillation

According to a report, OpenAI has claimed to have found evidence suggesting that Chinese AI company DeepSeek used OpenAI's proprietary models to train its own new model, DeepSeek-R1, through a technique known as "distillation". OpenAI stated that it had identified signs of such distillation, which involves extracting knowledge from a larger, more advanced model and transferring it to a smaller model, allowing for strong performance with lower computational costs.

The controversy arose when DeepSeek-R1 demonstrated capabilities in mathematics, programming, and reasoning that were seen as comparable to OpenAI's most advanced reasoning model, o1. However, DeepSeek-R1 was reportedly trained at only one-thirtieth of the cost of OpenAI's latest large model, drawing significant attention due to its cost efficiency and strong performance.

DeepSeek's technical report highlighted that DeepSeek-R1 surpassed several industry benchmarks, particularly excelling in mathematical and programming-related reasoning. This breakthrough positioned DeepSeek-R1 at the center of industry discussions, attracting interest from investors and analysts alike. However, its success also triggered a strong reaction from OpenAI.

OpenAI has taken a firm stance against unauthorized distillation, emphasizing that its service terms prohibit users from replicating its outputs or developing competing models based on its technology. OpenAI argues that DeepSeek's alleged use of distillation to extract knowledge from its models constitutes a clear violation of these terms.

While OpenAI has not publicly disclosed specific evidence, it confirmed that measures have been taken to restrict access to its API for certain accounts suspected of engaging in unauthorized activities. Sources indicate that OpenAI, in collaboration with Microsoft, is actively investigating accounts that utilized its API and has already revoked access for some users.

David Sacks, the White House's senior official overseeing AI and cryptocurrency, has publicly stated that there is substantial evidence of DeepSeek distilling knowledge from OpenAI's models and expressed strong disapproval of such practices. Additionally, U.S. government officials have launched a national security investigation into DeepSeek, citing concerns that its technology may involve intellectual property theft.

The dispute between OpenAI and DeepSeek underscores the escalating competition in the AI industry and the growing challenges surrounding intellectual property protection. As AI companies strive to safeguard their technologies from unauthorized use while ensuring innovation and accessibility, the debate over distillation techniques is expected to remain a critical issue.

Meanwhile, the U.S. government's stance on technological competition has sparked broader discussions, highlighting AI's increasing importance to national security. As AI continues to evolve rapidly, balancing innovation with protection will be a key policy challenge for global technology leaders.