Anthropic CEO Warns of AI Espionage, Calls for Government Action

As artificial intelligence continues to evolve at breakneck speed, so do the threats surrounding it. Dario Amodei, CEO of leading AI research company Anthropic, issued a stark warning this week: the most powerful AI algorithms in the U.S. may already be compromised by foreign espionage, with China as the primary suspect.

Speaking at an event hosted by the Council on Foreign Relations, Amodei claimed that Chinese actors are likely targeting top-tier U.S. AI labs in what he described as “large-scale industrial espionage.” These threats, he said, are not speculative. “Many of these algorithmic secrets... are $100 million secrets that are a few lines of code,” he explained. “I’m sure there are folks trying to steal them, and they may be succeeding.”

While Amodei didn’t provide evidence of specific breaches, the concern is real—and growing. The AI community is aware that some of the most valuable intellectual property in the world today lies in the algorithms and training techniques that drive frontier models. A few lines of source code, when properly understood, can represent years of research and hundreds of millions of dollars in investment.

In response, Amodei urged the U.S. government to step in and bolster defenses. While he didn’t outline exact measures during his talk, Anthropic recently submitted a set of policy recommendations to the White House Office of Science and Technology Policy (OSTP). These include closer collaboration between the federal government, U.S. intelligence agencies, and major AI labs to fortify security around sensitive research and development.

The comments underscore Amodei’s growing concern about China’s role in global AI development. In the past, he has openly supported stronger export restrictions on AI chips to China and warned about the use of AI for authoritarian purposes. He’s been particularly critical of DeepSeek, a Chinese AI model, citing poor results in a safety benchmark Anthropic conducted around bioweapons-related data.

Amodei’s remarks are not without controversy. Some experts argue that framing AI development as a zero-sum game between the U.S. and China could actually accelerate the kind of technological arms race many in the industry fear. Rather than collaboration to ensure safe global deployment, this perspective may push both nations toward secretive, aggressive development of ultra-powerful AI systems.

Indeed, critics of Amodei’s position have called for more open lines of communication and cooperation between U.S. and Chinese AI labs, emphasizing the existential risks of unchecked AI advancement. The argument is simple: if one nation builds an AI system that even its own scientists can't control, the consequences could be catastrophic—not just for one country, but for humanity.

Still, Amodei remains firm in his belief that the stakes are too high to leave security to chance. As frontier AI models grow more capable—and more valuable—he’s urging policymakers to act before it's too late. Whether Washington is ready to listen remains to be seen.

Cited: https://techcrunch.com/2025/03/12/anthropic-ceo-says-spies-are-after-100m-ai-secrets-in-a-few-lines-of-code/