Select Language:
OpenAI has alerted U.S. lawmakers that the Chinese AI startup DeepSeek is targeting ChatGPT’s developer and other leading American AI companies to steal models and repurpose them for their own training purposes, according to a memo obtained by Reuters.
Led by Sam Altman, OpenAI accused DeepSeek of “continual efforts to free-ride on the capabilities developed by OpenAI and other U.S. cutting-edge labs.”
The process they use, known as distillation, involves an older, more powerful AI system evaluating the responses generated by a newer model, effectively transferring the older model’s knowledge.
In a memo addressed to the U.S. House Select Committee on Strategic Competition with China last Thursday, OpenAI revealed: “We have observed accounts linked to DeepSeek employees developing methods to bypass OpenAI’s access controls and access models through hidden third-party routers and other obfuscation techniques.”
The memo also states: “We know that DeepSeek personnel created code to access U.S. AI models and extract outputs for distillation through automated means.”
Neither DeepSeek nor its parent company High-Flyer responded immediately to Reuters’ inquiries.
Based in Hangzhou, DeepSeek made waves early last year with a set of AI models that rivaled some of the top American offerings, raising concerns in Washington that China might catch up in AI development despite restrictions.
OpenAI claims Chinese large language models are “often cutting corners when it comes to safe training and deployment of new models.”
Previously, Silicon Valley leaders have praised models like DeepSeek-V3 and DeepSeek-R1, which are available worldwide.
OpenAI also said it actively removes users attempting to distill its models for rival development.
ChatGPT
Perplexity
Gemini AI
Grok AI
