President Donald Trump’s science and technology adviser on Thursday released a memorandum that says the government has found China is conducting industrial-scale campaigns to steal U.S. artificial intelligence technology.
Director of the White House Office of Science and Technology Policy Michael Kratsios sent a letter to the president explaining that foreign entities, primarily in China, are cracking into the United States’s networks to systematically extract frontier AI models.
Recommended Stories
Kratsios said that once the information is stolen, foreign actors are not able to replicate the full capabilities of U.S. AI, but rather create similar models to sell at a lower price.
“These distillation campaigns allow those actors to deliberately strip security protocols from the resulting models and undo mechanisms that ensure AI models are ideologically neutral and truth-seeking,” the memo said.
Kratsios said that while the U.S. is committed to equitable growth in AI technology, industrial distillation undermines American research, development, and information access, making it “unacceptable.”
“There is nothing innovative about systemically extracting and copying the innovations of American industry, and there is nothing open about supposedly open models that are derived from acts of malicious exploitation,” Kratsios said.
To tackle such threats, the president’s adviser said the administration will alert U.S. AI companies about the threat from foreign actors, allow private companies to coordinate against attacks and collaborate with them to develop best practices in defense, and explore measures to hold foreign entities accountable.
Anthropic is already rolling out technology that defends against industrial espionage, such as those that Kratsios described.
Earlier this month, the technology company launched a new cybersecurity initiative in a preview rollout with a select group of industry partners.
The effort, called Project Glasswing, uses Anthropic’s Claude Mythos frontier model to identify and exploit software vulnerabilities. The model is capable of performing a wide range of cybersecurity tasks, including identifying and exploiting vulnerabilities in code and testing systems for weaknesses.
“This is an important move from [Kratsios],” Sarah Heck, Anthropic’s head of public policy, said. “Industrial-scale distillation attacks are a serious national security threat. We need to have more of a spotlight on this issue, and look forward to continuing to collaborate on maintaining American AI leadership.”
MARYLAND IS EXPECTED TO BAN SURVEILLANCE PRICING. HERE’S WHAT THAT MEANS FOR CONSUMERS
Anthropic’s relationship with the federal government hasn’t been all that great, as it is engaged in a legal dispute with the War Department after the department scrapped a $200 million deal with the company.
But earlier this week, Trump signaled the tides could change when he told CNBC’s “Sqwauk Box” that Anthropic officials met at the White House. The president said he thinks Anthropic is “shaping up” and could be of “great use” to the administration.
