OpenAI Accuses DeepSeek of Illegally Training AI Model on U.S. Technology

OpenAI has accused Chinese AI firm DeepSeek of illegally using American AI models to train its own, raising concerns about economic and security risks.

In a formal memo to the U.S. Senate, OpenAI alleged that DeepSeek and other Chinese AI developers are exploiting illicit methods, specifically a technique called "distillation," to train their AI models. This involves training a new AI model based on the responses and outputs of a foundational model, in this case, OpenAI's ChatGPT.

According to OpenAI, DeepSeek employees have been accessing ChatGPT services using obscure third-party software and have developed software to programmatically extract outputs from American AI systems, according to a separate Reuters report.

OpenAI contends that these distillation techniques pose an economic threat to American companies. Unlike their U.S. counterparts, Chinese AI firms often do not charge subscription fees for their services. The company also asserts this poses a security risk, threatening U.S. dominance in the AI sector.

DeepSeek R1, which emerged last year, caused considerable concern among U.S. AI companies, offering comparable features at no cost and at a lower development cost.

Microsoft and OpenAI have been investigating whether DeepSeek utilized American models in its development. It remains unclear whether the memo to the Senate is a result of this investigation.

David Sacks, a White House AI advisor, voiced similar concerns nearly a year ago, stating that DeepSeek was extracting more performance from older chips, citing substantial evidence of distillation practices against OpenAI.