OpenAI has accused Chinese AI startup DeepSeek of using its proprietary models to train a competing AI system, according to Financial Times. The company claims to have found evidence suggesting that DeepSeek engaged in a practice called “distillation” to extract knowledge from OpenAI’s models, potentially violating its terms of service.
OpenAI alleges DeepSeek leveraged its models
Distillation is a common technique in AI development that allows smaller models to mimic the performance of larger, more advanced systems. However, OpenAI argues that using its outputs to create a rival product crosses the line. “The issue is when you are doing it to create your own model for your own purposes,” a person close to OpenAI told the Financial Times.
Microsoft and OpenAI reportedly investigated accounts believed to be linked to DeepSeek last autumn and subsequently blocked access due to suspected distillation activities. DeepSeek has not yet responded to the allegations, and OpenAI has not disclosed specific evidence to support its claims.
How DeepSeek caused the largest tech sell-off in history
US-China AI rivalry intensifies
DeepSeek has drawn significant attention in recent weeks with the release of its R1 AI model. Despite operating with far fewer resources than its American counterparts, DeepSeek R1 has achieved impressive performance benchmarks, positioning itself as a major competitor to OpenAI’s ChatGPT and Google’s Gemini.
Concerns over DeepSeek’s methods have coincided with volatility in the stock market. Nvidia shares tumbled 17% on Monday, erasing nearly $589 billion in market value, as investors speculated that DeepSeek’s cost-efficient model could reduce the need for high-end AI hardware. The stock rebounded by 9% on Tuesday.
David Sacks, an AI and crypto adviser in the Trump administration, commented on the situation, stating there was “substantial evidence” that DeepSeek had leveraged OpenAI’s outputs. However, he did not provide details on this evidence.
Is Nvidia stock still the AI king or losing its edge: The DeepSeek effect
OpenAI’s response
OpenAI reaffirmed its commitment to safeguarding its technology, stating, “We engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models.” The company also emphasized the importance of working with the US government to prevent AI technology from being appropriated by foreign competitors.
According to Financial Times, industry experts note that using the outputs of human-aligned commercial AI models to train new systems is a widespread practice. “It is not surprising that DeepSeek would be doing the same,” said Ritwik Gupta, a PhD candidate at the University of California, Berkeley. He pointed out that enforcing strict protections against such methods could prove difficult.
Meanwhile, the open-source nature of DeepSeek R1 has introduced new dynamics. Unlike ChatGPT, which operates as a closed system, DeepSeek R1 allows unrestricted access, potentially enabling other developers to build upon its foundation. Some analysts argue that this could pose further challenges for US-based AI firms in maintaining their technological edge.
Featured image credit: Jonathan Kemper/Unsplash
