TA419 Phishes AI Policy Experts With Microsoft AitM

How the Attack Worked

TA419, a China-aligned threat actor, ran a credential-phishing campaign against AI policy experts at U.S. think tanks, universities, and law firms. The operation relied on impersonation of prominent economists, AI policymakers, and an Anthropic employee to build credibility with targets. The email lure used the subject line "Request for Feedback on Military Integration of Claude" to draw in recipients working on AI policy issues.

The workflow was staged in two parts. First, attackers sent harmless outreach designed purely to establish trust and get a reply. Only after the target responded did the attacker send a shortened URL. That link triggered a multi-stage redirection chain, included a Cloudflare Turnstile bot check, and ultimately led to a fake OneDrive sign-in page built to look like a legitimate Microsoft login.

Why It Succeeded

The page used a frameless browser-in-the-browser technique combined with an adversary-in-the-middle proxy. This let the attacker capture both credentials and session cookies while still completing a real, successful sign-in to Microsoft services. Because the login worked as expected, victims had no visible indication that anything had gone wrong, since the usual signs of a failed or suspicious login were absent.

The layered approach, trust-building first, credential theft second, made the attack harder to spot than a typical one-shot phishing email. The bot check and multi-stage redirects also added a veneer of legitimacy that could make targets less suspicious before reaching the final page.

What to Watch For

• Unsolicited messages from unfamiliar contacts requesting expert feedback or commentary, especially ones that pivot quickly toward a link
• Shortened URLs that pass through several redirects before landing on a sign-in page
• Unexpected verification steps, like a bot check, inserted before a Microsoft or OneDrive login
• A sign-in flow that completes successfully but was reached through an unusual path

How to Build Resistance

Organizations and individuals who may be targeted by this kind of outreach, particularly those working on AI policy, defense, or regulatory issues, should treat unsolicited subject-matter requests with caution and verify the sender independently before replying or clicking any follow-up link.

Because AitM phishing can succeed even when the login appears normal, a successful sign-in should not be treated as proof that a page was legitimate. Adopting phishing-resistant authentication methods, such as passkeys, reduces the impact of stolen passwords and session cookies, since these methods are not vulnerable to the same credential-relay techniques used in this campaign.

Who did TA419 target in this campaign?

TA419 targeted AI policy experts at U.S. think tanks, as well as university researchers and legal professionals working on AI policy and regulatory matters.

How did the TA419 phishing attack work?

The attackers sent trust-building outreach impersonating economists, AI policymakers and an Anthropic employee, then followed up with a shortened URL that redirected victims through a Cloudflare Turnstile check to a fake OneDrive sign-in page using an adversary-in-the-middle technique.

Why didn't victims notice their credentials were stolen?

The AitM proxy let the sign-in event succeed normally while stealthily capturing credentials and session cookies, so there were no visible signs anything was wrong.

How can organizations defend against this kind of attack?

Awareness guidance includes verifying unsolicited outreach independently, being cautious of shortened links and unexpected bot checks before a Microsoft login, and enabling phishing-resistant authentication such as passkeys.