China-aligned TA419 impersonates Anthropic employee, ex-White House official to phish AI policy experts

What it is

TA419's attack kit uses Frameless BitB (browser-in-the-browser), an open-source tool that draws a fake Chrome login window inside the phishing page itself rather than opening a real browser popup — this is the specific mechanism the existing summary names but doesn't explain.

How it works

flowchart TD
    A[Benign invitation email] -->|target replies| B[Shortened link sent]
    B --> C[Domain 1: Cloudflare Turnstile + fake OneDrive loading screen]
    C --> D[Domain 2: fake OneDrive folder listing]
    D -->|click document or sign-in prompt| E[Frameless BitB renders fake Chrome window]
    E --> F[Credentials + MFA relayed live to real Microsoft login]
    F --> G[Custom script auto-accepts 'Keep me signed in', submits OTP codes]
    G --> H[Session cookie captured → account access]

The relay is live and adversary-in-the-middle, not a static credential harvester: entered passwords and MFA/conditional-access checks pass through to Microsoft's actual servers in real time, so the login genuinely succeeds while attackers watch and capture the resulting session cookie. A custom script embedded in the kit handles the "Keep me signed in" prompt and submits one-time codes as soon as the victim's MFA validates — this automation is what lets the attackers bypass MFA without needing to phish the code themselves.

Preconditions / Affected

Targets must reply to the initial lure email (no malicious link is present until that point) and then complete a Microsoft/OneDrive-style cloud login on the spoofed page. The campaign specifically targets US AI policy experts, think-tank analysts, and figures with foreign-policy/export-control expertise.

What to do

The page gives no patch or product fix (this is a social-engineering/credential-phishing campaign, not a vulnerability), but it does supply concrete detection/threat-intel indicators:

• Domains are registered via NameSilo and fronted by Cloudflare CDN to hide backend hosting IPs — worth flagging in email/proxy logs.
• Known impersonation domains cited: heritiages[.]org, heritiage[.]org (Heritage Foundation lookalikes), tw-koryu[.]org (Japan-Taiwan Exchange Association), shinjirou[.]info (Japanese Defense Minister Shinjirō Koizumi lookalike).
• Domain themes consistently mimic file-sharing/cloud services (e.g., OneDrive) — a useful filter for mail-gateway rules.
• Because MFA and conditional-access checks are relayed live, standard MFA is not sufficient defense against this kit; phishing-resistant authentication (e.g., FIDO2/hardware keys) would not be bypassable the same way, though the page itself does not state this recommendation explicitly.

Caveats

Proofpoint's attribution of intent ("likely supports wider Chinese intelligence objectives") is an assessment, not confirmed; the page does not disclose how many targets were compromised or whether any of the named lookalike domains are still active.