Chinese Hackers Impersonated Anthropic to Steal AI Secrets

Proofpoint says operatives impersonated a senior Anthropic employee and emailed a think-tank researcher with the subject 'Request for feedback on military integration of Claude'.

In a new report from the cybersecurity firm Proofpoint, a hacking group identified as TA419 and attributed to China used forged identities — including Anthropic researchers and former White House staffers — to solicit sensitive information about artificial intelligence development in the United States. Proofpoint says the campaign targeted AI policy specialists at think tanks, universities and law firms.

Scope and methods

According to the report, the attackers began with legitimate-looking, seemingly harmless messages inviting recipients to join a fictional 'AI policy advisory committee' or to participate in drafting a fabricated report for the Senate Foreign Relations Committee. After a short exchange of messages, the operators delivered files containing malware designed to gain access to the targets' cloud accounts.

Proofpoint highlights one instance in which the group, while posing as a senior Anthropic staffer, sent the message with the subject line 'Request for feedback on military integration of Claude' to a policy researcher. The company says the impersonations also extended to claiming former White House affiliations to improve credibility.

Attribution and policy context

Mark Kelly, a researcher at Proofpoint, told CNN that experts are confident TA419 is a Beijing-aligned threat actor. Proofpoint says that conclusion rests on the consistent alignment of targets with Chinese government interests, observed infrastructure and technical indicators, and confirmations from industry partners.

The report arrives amid a fraught US debate over AI oversight. Proofpoint notes public and industry calls for regulation — a position echoed by Anthropic CEO Dario Amodei, who has warned about uncontrollable risks from the technology. At the same time, figures including Donald Trump, Mark Zuckerberg and Jensen Huang have challenged regulatory approaches. A US convening of industry leaders produced a document described as an 'ethically binding' agreement, which Donald Trump said would enable self-regulation; proponents of a more permissive approach argue the United States should not cede an AI arms-race advantage to China.

Proofpoint warned TA419 will likely continue targeting think tanks and AI policy experts while continuing to impersonate legitimate domain authorities.