PROOFPOINT researchers have uncovered a China-aligned TA419 credential-phishing operation that targeted United States artificial intelligence policy experts, think tanks, and universities in July 2026. The attackers conducted multi-stage social engineering, beginning with benign outreach that invited targets to join fictitious AI policy advisory bodies or contribute to congressional reports on semiconductor supply chains.
To build trust, they impersonated well‑known public figures, such as Lynne Parker of the White House OSTP and Heidi Crebo-Rediker, before delivering a malicious link. The campaign used a browser-based credential‑phishing kit to direct victims through shorteners to an intermediate validation gate, then to a fake OneDrive interface protected by a Cloudflare Turnstile challenge, before landing on an Adversary-in-the-Middle phishing portal.
The attackers employed Frameless BitB, an open‑source tool that creates a deceptive browser window within the page while proxying live traffic to Microsoft authentication servers. Custom JavaScript was injected to monitor inputs and capture session cookies, enabling the actors to complete sign‑ins and capture credentials, MFA codes, and session tokens behind the scenes.
The operation shows a pattern of domain registration via NameSilo, hidden hosting behind CDNs, self‑signed certificates, and routing through residential proxies to mask locations. A prior February 2026 wave spoofed Anthropic personnel to discuss Claude, indicating broader attempts to expose Western AI policy discussions.
Defence implications call for stronger origin‑bound credentials and phishing‑resistant options, alongside verification via secondary channels and ongoing staff training to recognise multi‑stage lures.