AI Policy Invitations Hid a Microsoft 365 Session Trap

A China-aligned group tracked as TA419 tried to enter the cloud accounts of people shaping artificial-intelligence policy by exploiting professional trust rather than a software flaw. Proofpoint published the previously undisclosed activity on October 1, after observing campaigns against specialists at US think tanks, universities and law firms.

The first messages looked like ordinary professional outreach. In July, senders impersonated former White House technology official Lynne Parker and economist Heidi Crebo-Rediker, inviting recipients to join a fictitious AI policy committee or contribute to a supposed Senate report on export controls. A February campaign had impersonated an Anthropic employee.

How the sign-in trap worked

After a recipient replied, the attacker sent a shortened link. It passed through actor-controlled domains, displayed a fake OneDrive loading screen and used a Cloudflare Turnstile check before opening an adversary-in-the-middle phishing page.

The page relayed a genuine Microsoft 365 authorization flow while placing a browser-like overlay above it. That design could collect the password and one-time authentication code, then steal the resulting session cookie. Proofpoint said custom scripts watched the victim's progress and automatically submitted validated codes. This matters because a stolen authenticated session can bypass the protection that a password and conventional multi-factor authentication normally provide.

The campaign targeted the relationships and private discussions around AI regulation, export controls and national strategy, not an AI model itself.

A narrow campaign with wider implications

Reuters identified one recipient, Alex Engler of the Penn Center on Media, Technology, and Democracy. He noticed that the invitation felt wrong and checked through other contacts. Proofpoint said the campaign involved fewer than ten people at a handful of organizations, suggesting focused intelligence collection rather than broad credential theft.

The disclosure also shows why sensitive technology policy can attract espionage even when no source code is involved. Access to email and cloud documents may reveal unpublished policy positions, professional networks and export-control discussions. That context connects to NewTqnia's coverage of South Korea's expanded espionage law and its emphasis on protecting strategic technology information.

For likely targets, Proofpoint recommends verifying unexpected outreach through another channel and using phishing-resistant, origin-bound authentication such as passkeys. The lesson is not that every professional invitation is malicious, but that familiar names and genuine sign-in pages no longer prove that the path leading to them is trustworthy.