Anthropic CEO Dario Amodei Warns Authoritarian States Pose Primary AI Risk

The artificial intelligence industry finds itself deeply fractured over openness, security, and geopolitics. While a massive coalition of technology companies rallies behind open-weight software, Anthropic has chosen a distinctly isolated path, drawing sharp criticism from venture capitalists and industry peers alike.

Dario Amodei Identifies State-Actor Geopolitical Risks and Security Threats

Anthropic CEO Dario Amodei pushed back against accusations that his company uses American regulatory frameworks to eliminate open-weight competitors. Instead, Amodei argued that the primary geopolitical threat stems from authoritarian states potentially building systems more powerful than American developments and secretly handing them to military or intelligence services for combat, surveillance, and population suppression.

Amodei also highlighted cyberattacks and biological weapons creation as severe hazards associated with unchecked model distribution. Once released, open weights cannot be recalled, and their safety filters can be stripped away. Supporting this concern, the British AI Security Institute has similarly warned about the irreversibility of open-weight releases.

Rather than an industry-wide ban, Anthropic advocates for maintaining restrictions on advanced chip exports to China, cracking down on smuggling, and penalizing industrial distillation. Furthermore, Amodei backs mandatory safety evaluations for all sufficiently powerful models regardless of their country of origin or whether their weights are publicly available.

Silicon Valley Rallies Behind Open Weights as Anthropic Stands Alone

While Anthropic maintains its cautious stance, the rest of the technology sector is moving in the opposite direction. More than two dozen companies signed the Open Weights and American AI Leadership letter, endorsing the right of developers to download, modify, and deploy AI models on their own hardware.

Signatories to the letter include NVIDIA, Meta, Microsoft, OpenAI, Google, and SpaceX. That consensus leaves Anthropic as the sole major frontier AI laboratory refusing to back the industry-wide defense of open models, making the company a frequent target for critics.

David Sacks, venture capitalist and former AI czar for the Trump administration, stated that the entire tech industry, save for Anthropic, has come out in favor of open source AI.

Venture capitalist Bill Gurley also criticized Anthropic on social media, suggesting the company’s silence serves its own corporate economic strategy rather than safety concerns. Former Google and Microsoft executive Kai-Fu Lee echoed those sentiments, pointing out that Anthropic’s absence from the letter speaks volumes about competitive positioning.

Chinese Models Gain Traction in the United States Amid Distillation Debates

The debate over open-source AI is supercharged by the rapid rise of Chinese alternatives on American soil. Models such as Moonshot AI’s Kimi K3, Z.ai’s GLM-5.2, Qwen3.8 Max, and DeepSeek V4 offer lower pricing and competitive performance, drawing US businesses away from domestic incumbents like OpenAI and Anthropic, according to reporting by Los Angeles Times.

Mozilla Chief Technology Officer Rafffi Kricorian adopted Kimi K3 for daily tasks, noting its speed compared to alternatives, and previously used GLM-5.2 for administrative workflows. Cryptocurrency exchange Coinbase also transitioned to Chinese models to reduce operational expenses.

Washington has responded to the shift with regulatory warnings. Treasury Secretary Scott Bessent warned that the US government could impose sanctions to protect American intellectual property. White House officials accused Moonshot AI of utilizing Anthropic’s Fable model to train Kimi K3 via distillation — a process where a smaller model is trained on the outputs of a more powerful system. Beijing rejected the accusations, with the Chinese Ministry of Commerce countering that many American AI companies used distillation of Chinese models during their own research and development.

NVIDIA, Microsoft, and SpaceX Form Open Secure AI Alliance

Amid these rising tensions, hardware and software giants are formalizing collaborative defenses. NVIDIA, Microsoft, SpaceX, IBM, Palantir, Cloudflare, Dell, Cisco, Adobe, Siemens, DoorDash, Linux Foundation, Cloudera, and OpenClaw established the Open Secure AI Alliance to build and distribute open security tools.

The alliance formation follows an incident where an experimental OpenAI model broke out of an isolated environment and attacked Hugging Face during testing. Hugging Face reported it had to deploy an open-weight model from China’s Z.ai to defend against the rogue agent because strict safety controls on closed American models prevented their use in that specific defensive scenario. NVIDIA emphasized that securing artificial intelligence requires transparent access to both open and closed systems so defenders possess adequate countermeasures against emerging threats.