OpenAI has accused DeepSeek of malpractice in developing the next version of its artificial intelligence model — even before any official launch.
“DeepSeek’s next model (whatever its form) should be understood in the context of its ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs,” OpenAI said in a memo to the U.S. House Select Committee on China on February 12.
DeepSeek has not confirmed any new launches.
OpenAI’s concerns may stem from expectations that the Hangzhou-based company could make a major announcement during next week’s Lunar New Year celebrations, echoing the surprise rollout it staged last year.
“In this high-stakes AI arms race, there is a constant fight on the knowledge, capability, accuracy, and efficiency of the models,” Neil Shah, vice president of research at Counterpoint Research, told Rest of World. The conflict reflects larger tensions in the global development of AI, Shah said.
DeepSeek became a frontrunner in the global AI race almost overnight when it released its R1 model during the Lunar New Year period last year, claiming performance comparable to top U.S. models despite being trained with far fewer advanced chips. The launch reignited debates in Washington over whether U.S. export controls on advanced semiconductors were sufficient to preserve America’s AI lead.
In its memo, OpenAI accused DeepSeek of using “distillation” techniques — a common method where a smaller model is trained on the outputs of a more powerful one to replicate its capabilities.
“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 said. “We also know that DeepSeek employees developed code to access U.S. AI models and obtain outputs for distillation in programmatic ways. We believe that DeepSeek also uses third-party routers to access frontier models from other U.S. labs.”
OpenAI said it does not permit its outputs to be used to create “imitation frontier AI models” that replicate its capabilities.
This is not the first time OpenAI has raised concerns about distillation. The company investigated whether DeepSeek had distilled its data right after the R1 model launched in January last year.
Austin Horng-En Wang, an associate political scientist at the think tank RAND Corporation, questioned why OpenAI chose to escalate now, particularly as Chinese companies and policymakers have openly pushed for an open-source AI ecosystem since DeepSeek’s breakthrough last year.
“One possible reason for the accusation is to prevent DeepSeek and China companies from acquiring more chips to distill the U.S. model, so that the U.S. models can keep their leading position,” Wang told Rest of World.
DeepSeek’s R1 model has helped spur China’s embrace of open-weight AI models — systems that developers worldwide can download, modify, and deploy. That approach contrasts with the closed systems favored by most U.S. tech giants, which tightly control access to their models, data, and architecture.
In the past month, Chinese tech giants and AI startups have rushed to release their latest open models ahead of DeepSeek.
“The reality is none of the models is an island and the entire industry has mostly evolved based on recursive learning,” Shah said. “The newer entrants are in many instances going through the same routes of ‘distillation’ and ‘optimization.’”
