Anthropic’s CEO Dario Amodei is raising concerns about spies, likely from China, getting access to valuable “algorithmic secrets” from top AI companies in the U.S. He is urging the U.S. government to take action.

Speaking at a Council on Foreign Relations event, Amodei highlighted China’s reputation for engaging in large-scale industrial espionage, emphasizing that companies like Anthropic are likely targets.

According to Amodei, these algorithmic secrets can be worth millions of dollars and are just a few lines of code. He believes there are individuals attempting to steal them, potentially with success.

Amodei stressed the importance of increased support from the U.S. government to address this threat, although he did not specify the exact nature of assistance required.

While Anthropic did not comment specifically on Amodei’s remarks to TechCrunch, the company did refer to its recent recommendations to the White House’s Office of Science and Technology Policy (OSTP).

In its submission, Anthropic proposed that the federal government collaborate with AI industry leaders to enhance security at cutting-edge AI labs, including working closely with U.S. intelligence agencies and their international partners.

Amodei’s concerns stem from his critical views on Chinese AI development, advocating for stricter U.S. export controls on AI chips to China. He also highlighted DeepSeek’s poor performance on a bioweapons data safety test conducted by Anthropic.

Amodei has emphasized in his writings, such as the essay “Machines of Loving Grace,” the risks associated with China’s use of AI for authoritarian and military purposes.

While some critics within the AI community argue for increased collaboration between the U.S. and China to prevent an arms race resulting in uncontrollable AI systems, Amodei stands firm on the need for caution in sharing advanced AI technology.