OpenAI accuses DeepSeek of model distillation using unauthorized access
OpenAI accused Chinese AI startup DeepSeek of using unauthorized techniques to distill ChatGPT and other US AI models by circumventing access restrictions and programmatically extracting model outputs for training its own systems. DeepSeek and its parent company High-Flyer have not responded to the allegation.
What happened
OpenAI submitted a memo to the US House Select Committee on Strategic Competition on 12 February 2026 accusing DeepSeek of targeting OpenAI and other US AI companies to replicate their models using distillation techniques.
OpenAI stated that accounts associated with DeepSeek employees developed methods to circumvent OpenAI's access restrictions using obfuscated third-party routers and other masking techniques, and created programmatic code to access US AI model outputs at scale for distillation purposes.
Distillation involves using an older, more established and powerful AI model to evaluate the quality of answers from a newer model, effectively transferring the older model's knowledge to the new model.
Reported impact
OpenAI stated that Chinese large language models are "actively cutting corners when it comes to safely training and deploying new models."
The accusation raises concerns about China catching up in the AI race despite US restrictions, following DeepSeek's earlier disruption of global markets with its R1 and V3 models.
Index analysis
While distillation as a technique is not inherently illegal, obtaining model outputs without authorization may violate service terms and potentially trigger legal liability.
OpenAI stated it proactively removes users who appear to be attempting to distill its models to develop competing models.
Governance controls
OpenAI said it maintains access restrictions and has procedures to identify and remove accounts engaged in unauthorized distillation.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.