OpenAIWARNsDeepSeek Uses Distillation to Leverage US Models (NASDAQ:OPEN, NASDAQ:ANTH)

OpenAI notified the House Select Committee on China that its rival DeepSeek is leveraging distillation—using outputs from OpenAI and other US models to train new capabilities—via increasingly obfuscated methods to evade its platform controls. The memo, dated February 12, 2026, follows a probe into whether DeepSeek improperly accessed data after the R1 model’s release in 2025. DeepSeek reportedly circumvented guardrails by routing through third-party systems, accessing models programmatically, and using unauthorized resellers. These practices could erode US AI leadership and threaten OpenAI and Anthropic, which charge for premium services. OpenAI also highlighted risks from censored content enabling broader misuse in high-impact areas like biotech and chemistry. The company said DeepSeek employees are developing agent-based models, with a potential release targeted for late 2025. US-China tensions over AI and semiconductors continue, with access to advanced chips potentially accelerating DeepSeek’s progress despite export restrictions on certain hardware.