Anthropic CEO Warns of DeepSeek’s AI Risks as Chinese Model Gains Traction in Silicon Valley

CEO Dario Amodei from Anthropic showed concern about DeepSeek because it was acquiring more power in Silicon Valley through its R1 model. Anthropic CEO Dario Amodei directs his focus toward both national privacy intrusions and security weaknesses that face the country.

Why It Matters

Analyzing Anthropic DeepSeek for safety purposes yielded the weakest test results ever since the interview with Amodei for ChinaTalk podcast about bioweapons-related information. The model has no programming features that limit such output generation according to his ChinaTalk interview.

Security Concerns

Anthropic systematically tests its AI models to determine their capability to generate bioweapon-related information which exists beyond the scope of public internet databases. Amodei declared that DeepSeek's current models do not pose genuine danger but he predicted their threat profile to increase in the near future. DeepSeek needed to prioritize AI safety according to his opinion.

The defense expert stands behind current U.S. regulations that restrict AI chip exports to China because he feels such advanced technology would boost Chinese military power.

Broader Implications

Amodei did not specify which DeepSeek model was tested or provide further technical details. Neither Anthropic nor DeepSeek responded to TechCrunch’s request for comment.

Concerns about DeepSeek’s safety extend beyond Anthropic. Cisco security researchers recently reported that DeepSeek R1 had a 100% jailbreak success rate in safety tests, failing to block harmful prompts on cybercrime and illegal activities. While bioweapons were not mentioned, Meta’s Llama-3.1-405B and OpenAI’s GPT-4o also had high failure rates of 96% and 86%, respectively.

Market Impact

Despite safety concerns, DeepSeek’s adoption continues to grow, with AWS and Microsoft integrating R1 into their cloud platforms—even as Amazon remains Anthropic’s largest investor. Meanwhile, organizations like the U.S. Navy and the Pentagon have begun banning DeepSeek over security concerns.

The Competitive Landscape

Amodei acknowledged DeepSeek as a rising competitor, placing it alongside top U.S. AI firms such as Anthropic, OpenAI, Google, Meta, and xAI. "The new fact here is that there’s a new competitor," he said. "DeepSeek is maybe being added to that category."