OpenAI and Anthropic raise alarm over China's Deepseek in warnings to US government

The company describes Deepseek as "simultaneously state-subsidized, state-controlled, and freely available," arguing this combination threatens both individual privacy and intellectual property rights.

Anthropic's filing focuses on biosecurity concerns, particularly Deepseek-R1's willingness to provide information about biological weapons, even when users express harmful intent. While not an immediate threat, this lack of safety measures demonstrates the need for better government oversight of AI systems, Anthropic argues.

Neither company addresses how open-source models like Deepseek-R1 could threaten their business models, which rely on selling access to proprietary AI systems and interfaces. Microsoft CEO Satya Nadella recently claimed that AI models "are getting commoditized."

Anthropic reveals potential regulatory gap in US chip restrictions

Anthropic makes an interesting point about what it sees as a critical oversight in US export restrictions on AI chips to China. While Nvidia's H20 chips meet the reduced performance requirements for Chinese export, they "excel at text generation (sampling)" which is a "fundamental component" of reinforcement learning. Since this approach drives recent advances in reasoning models like Deepseek-R1, Anthropic urges immediate regulatory action.

Both companies acknowledge America's shrinking technological advantage. "While America maintains a lead on AI today, DeepSeek shows that our lead is not wide and is narrowing," OpenAI writes.

Google's response focuses on copyright and fair use, while raising concerns about how new AI export rules could disadvantage U.S. cloud providers. Unlike its competitors, Google doesn't mention Deepseek in its statement.