What Readers Need to Know

• Former OpenAI and Anthropic researcher Jacob Coxon says he is 90% certain Chinese spies are already inside OpenAI and Anthropic, but he has publicly offered no evidence proving the allegation.
• The warning lands after the Trump administration explicitly directed federal agencies to work with leading AI developers against “malicious cyber actors” and “insider threats.”
• If an AI company knowingly ignored serious security weaknesses, the legal consequences could extend beyond stolen trade secrets to export controls, federal contracting obligations and national-security scrutiny.

By Samuel López | USA Herald

SAN FRANCISCO, Calif. - Former Anthropic researcher Jacob Coxon has made an extraordinary allegation about the security of America's most valuable artificial intelligence companies — and, if there is evidence behind it, the question is no longer simply whether China is stealing American technology.

It is whether American AI companies are doing enough to stop it.

Coxon, who has worked at both Anthropic and OpenAI, told entrepreneur Patrick Bet-David that he believes Chinese intelligence has already penetrated the companies.

“I think one really important thing people don't realize is that China already has spies at OpenAI and Anthropic,” Coxon said, adding that he was 90% positive. “It's almost certain that they have spies already there … There's no serious background check for this sort of stuff yet.”

South China Morning Post independently reported Coxon's allegation this week and noted that he provided no public evidence identifying an alleged spy or demonstrating that either company has been infiltrated. That distinction matters. A former employee's 90% confidence is not proof. South China Morning Post

But the underlying threat is hardly hypothetical.

In January, a federal jury convicted former Google engineer Linwei Ding of seven counts of economic espionage and seven counts of trade-secret theft after prosecutors proved he stole more than 2,000 pages of confidential Google AI information while secretly associating with Chinese technology ventures. Department of Justice

Coxon's larger argument is that America's technological lead can disappear through espionage.

“China has the ability to hack at any time,” he said. “If we pull ahead, that doesn't give us a big lead. They can just step in, steal our progress at any time and catch up.”

He went further.

“If we slow down, it will probably slow down China quite a bit, too, because, currently, a lot of their progress is from essentially copying or stealing U.S. work — not all of it, but there is a component.”

And Coxon believes Washington may eventually have to stop treating frontier AI like ordinary intellectual property.

“This is a super weapon,” he said. “So soon people are going to have to realize that this is not like a casual business thing.”

That warning collides directly with Trump administration policy.

Trump's 2025 AI Action Plan states that advanced American technology must be protected from being “misused or stolen by malicious actors.” More specifically, the plan calls on the Defense Department, Homeland Security, Commerce Department and intelligence community to work with leading AI developers to protect AI innovations against malicious cyber actors and insider threats. The White House

So, what happens legally if Coxon is right?

There is no general federal law automatically making an AI company's management criminally liable simply because an employee turns out to be a foreign intelligence operative. Nor does every frontier-AI employee require a government security clearance.

But companies cannot simply ignore security.

Federal trade-secret law defines protected information partly by whether its owner took “reasonable measures” to keep it secret. Weak access controls, uncontrolled downloading or ignored insider-risk warnings can therefore become legally significant — and potentially weaken the company's own claim that stolen information qualified as a protected trade secret. Legal Information Institute

The stakes climb further where export-controlled technology, classified information or federal contract information is involved. Government contractors can face specific cybersecurity and access-control requirements, while Defense Department contractors handling covered information must provide “adequate security.” Acquisition.gov

And Washington is already treating AI theft as a national-security matter. Economic espionage intended to benefit a foreign government is prosecuted under 18 U.S.C. §1831. Legal Information Institute

Coxon's allegation remains unproven. But the Google prosecution demonstrates why it cannot simply be laughed off.

The uncomfortable question for America's AI giants is becoming harder to avoid: if the technology is important enough for Washington to protect like a strategic national asset, are companies still protecting it like ordinary corporate intellectual property?

About the Author

Samuel López is a reporter and news correspondent for USA Herald and a veteran legal analyst with more than two decades of experience examining state and federal litigation, regulatory disputes, insurance matters and complex legal issues. His reporting increasingly focuses on the collision between artificial intelligence, corporate liability, government regulation and national security.

Want the stories behind the headlines? Get USA Herald's investigative reporting, legal analysis and national-security coverage delivered directly to you through the USA Herald newsletter. Follow USA Herald on X for breaking developments and new reports at @RealUSAHerald.