Dario Amodei on China's Open-Source AI and Looming Cyber Risks

Overview

Dario Amodei argues that highly capable AI models have a major advantage: users generally prefer greater intelligence, so the most valuable uses are often found at the frontier. He sees a thriving role for less capable models, but worries that powerful cyber capabilities could eventually spread through downloadable models.

Amodei says models with “Mythos-class” cyber abilities may be available to anyone within about a year. He hopes systems will be patched by then, but does not think their release can be prevented.

China and AI

Amodei says his year working at Baidu taught him about speech recognition. He recalls that a remark about privacy not being a concern in China—and the company having abundant speech data—stood out to him.

His larger concerns about China are geopolitical. He points to authoritarian influence and the suppression of criticism, including pressure reaching into US business networks.

AI’s Potential Impact

Amodei warns that AI could strengthen authoritarian control and create a dystopian future. But he says it could also support democracy, freedom, and equal justice. Which path it takes, he argues, depends on AI companies, governments, and the public.

Presenters and Contributors

• Emily Chang (interviewer)
• Dario Amodei (Anthropic CEO)