IncidentAttacks on the AI sector
TA419: China-aligned phishing posed as an Anthropic employee and US officials to reach AI policy experts
Proofpoint reported that TA419, a China-aligned espionage group, has run credential phishing against AI policy experts at US think tanks, universities and law firms since at least April 2025, activity it says had not been reported before. In July 2026 the group impersonated a former White House science official and a foreign policy economist, with invitations to a fictitious AI policy advisory committee and a Senate report on AI export controls; in February 2026 it posed as a senior Anthropic employee asking a think tank analyst for feedback on military uses of Claude. The links led to an adversary-in-the-middle page that captured Microsoft 365 sign-ins, including the MFA step. Proofpoint confirmed no compromise.
Replay the attack step by step in the Learning Center
How it happened
- Emails impersonated real AI policy figures, with lures about advisory committees, export controls and the military integration of Claude.
- Shortened links redirected through several sites to a fake Microsoft OneDrive page built with a customized Frameless BitB overlay and an Evilginx phishlet.
- The page relayed the login to Microsoft 365 in real time, capturing the password, the session and the MFA approval.
Impact
The people who shape AI policy, and the organizations that employ them, were targeted for their accounts and correspondence; no compromise was confirmed.
The fix
Phishing-resistant, origin-bound sign-in (passkeys or security keys), which a relay page cannot replay, and verifying unsolicited expert outreach through an independent channel.
Lessons
- Interest in AI policy is itself a lure: an unsolicited invitation to advise is a pretext stage.
- One-time codes and push approvals do not stop a relay page; only origin-bound credentials do.