OpenAI escalates DeepSeek distillation battle with government complaint
OpenAI has flagged the Chinese AI brand, DeepSeek to the US government, alleging the startup used distillation to train its AI on OpenAI's models.
OpenAI has taken complaints about the Chinese artificial intelligence startup DeepSeek to the US government, claiming that the company has used its older large language models to train its own AI models.
OpenAI alleged in a note sent to the U.S. House Select Committee on Strategic Competition that DeepSeek was targeting its technology and those from similar US companies to use as a base for its AI training.
The process, called distillation, uses an older more powerful AI model to fact check the answers produced from a newer model. The newer model then obtains the information of the older model without having to train on any data.
"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," OpenAI said in the note, viewed by Reuters.
"We also know that DeepSeek employees developed code to access U.S. AI models and obtain outputs for distillation in programmatic ways," the memo added.
DeepSeek became an industry darling in early 2025 amid the launch of its DeepSeek-V3 and DeepSeek-R1 models. The company was praised for bringing forth an open-source generative-AI option that used less powerful GPUs for training. Several AI associated brands saw massive drops in market value during that time as the industry was revving up to invest even more money into high powered AI GPUs. However, DeepSeek has been marred by claims of unauthorized “reverse engineering” of other popular models to train its chatbot.
Despite OpenAI's efforts to protect its technologies, DeepSeek has maintained its place as a competitor and remains available globally. So far, OpenAI said its primary mode of action against the company has been to remove users that seem like they are trying to execute distillation methods on its models to extract data for outside development, Reuters noted.