Chinese Hackers Borrowed Former White House Science and Tech Expert’s Name to Phish Inboxes Writing America’s AI Export Rules
La boîte aux lettres qui rédige une phrase de contrôle d'exportation est un prix plus silencieux qu'une pile de poids de modèle, et c'est la boîte aux lettres qu'un équipage aligné sur la Chine a poursuivi. À compter du 8 juillet 2026, Proofpoint later reported, un groupe qu'il suit comme TA419 a ouvert l'hameçonnage des titres de compétence en empruntant Lynne Edwards Parker, décrit là comme l'ancien directeur adjoint principal du Bureau de la Maison Blanche Politique de la science et de la technologie. La même course a ensuite emprunté Heidi Crebo-Rediker, un économiste dont la carrière est assurée par le Département d'État et le Comité sénatorial des relations extérieures. Les notes ne demandaient pas de code source. Ils ont demandé aux responsables politiques de se joindre à une conversation.
Proofpoint appelle TA419 espionnage-motivé et Chine-aligné, et dit que la rédaction du 1er octobre 2026 a été le premier rapport public sur le groupe. Depuis au moins avril 2025, elle observe des phishs de TA419 aux États-Unis et au Japon, des groupes de réflexion, des entrepreneurs de défense, des universités et des cabinets d'avocats. Reuters reported the same day que la vague de juillet a touché moins de 10 personnes, y compris des spécialistes de la réglementation de l'IA, du contrôle des exportations et de la stratégie nationale. Le label Chine est l'évaluation de Proofpoint à partir de l'infrastructure et le ciblage. Reuters a noté que l'ambassade de Chine n'a pas commenté et que Pékin nie le cyberespionnage. Ce n'est pas une décision de justice.
Why the borrowed biography had to match the ask
The July messages were written to sound like a favor between colleagues. One version invited the recipient onto a fictitious « AI Policy Advisory Committee. » Another asked for help with a Senate Committee on Foreign Relations report on AI export controls and supply chains. Help Net Security relayé le compte de Mark Kelly que l'email d'ouverture ne portait aucun lien malveillant. L'appât était la réponse. Ce n'est qu'après l'arrivée d'une adresse web raccourcie que quelqu'un a écrit en promettant plus de matériel et en passant par les redirections vers une connexion cloud contrefaite.
Crebo-Rediker’s record is why the second name fit. She was the State Department’s first chief economist in the Obama years, had been chief of international finance for the Senate Foreign Relations Committee, and is a senior fellow at the Council on Foreign Relations. A request to help with a Senate paper on export controls is a note that career would generate. Parker’s name did the parallel job for an advisory committee. She is a roboticist, founding director of the National Artificial Intelligence Initiative Office, and, in Proofpoint’s wording, a former principal deputy director of the science office.
The Trump administration had already advertised the persona. On 22 December 2024, Donald J. Trump named Parker executive director of the President’s Council of Advisors on Science and Technology and counselor to the director of the Office of Science and Technology Policy. Michael Kratsios was to run that office, and David Sacks, whom Trump called an « AI and Crypto Czar, » was to chair the council. The announcement, carried by Politico et described by Science, promised a « Golden Age of American Innovation » and « America’s technological dominance, » stapling a science panel to a cryptocurrency pitch. That branding handed a later impostor a committee title that already sounded official. The people drafting export-control language were still working in ordinary Microsoft 365 mail. The White House sold the names and left the inboxes easy to phish.
Heidi Crebo-Rediker’s State Department and Senate Foreign Relations background matched the AI export-control report lure.
Reuters identified one recipient as Alex Engler, a former White House official who now heads the University of Pennsylvania’s Center on Media, Technology, and Democracy. Engler said the message, which invited him « to join a new AI policy project, » « felt slightly, nebulously off. » He realized it was an impostor after checking with other people. Parker told Reuters he was one of two people she knew of who had received mail purporting to come from her in early July. She also said, « The United States and China are in a competition around AI. Trying to get people in the AI policy space to reveal information about their AI policy plans — if that indeed was what the objective was — it’s not surprising. »
A small list is the point, not a consolation. Criminal kits spray tens of thousands of inboxes and still function as a business. In December 2025, The AEGIS Alliance reported on the arrest tied to RaccoonO365, a service built to steal Microsoft 365 sessions at volume. TA419 aimed the same class of theft at a room that fits around one table. The target was not a model file. It was the thread where an export rule is still being worded.
What the fake OneDrive page took from a Microsoft 365 login
Proofpoint’s account, written for defenders and not as a build guide, describes two stages. The first July domain, driftshare[.]co, showed a fake OneDrive screen behind a Cloudflare Turnstile check. A second domain, globalfileshareplatform[.]com, hosted the adversary-in-the-middle page, using a customized browser-in-the-browser kit known as Frameless BitB and an Evilginx phishlet for Microsoft 365. The page relayed the real Microsoft sign-in, including the one-time code, so the password and the second factor were typed where an operator could watch.
Scripts on the page drew a file listing and a fake browser window, accepted « Keep me signed in » without the user, and submitted a one-time code after it checked out. The login still succeeded at Microsoft. What the operator kept was the password, the multi-factor code, and the session cookies that mark a Microsoft 365 user as already signed in. Microsoft has described that pattern for years. A copied cookie means the second factor is not asked again.
Proofpoint described a counterfeit browser layered over a OneDrive-style share that relayed the real Microsoft sign-in.
That 16 July 2026 talk by Evilginx developer Kuba Gretzky covers credential-relay kits and attempts to weaken phishing-resistant sign-in. It is not a briefing on TA419. It shows the same family of tool Proofpoint says this crew customized.
Domains sat behind Cloudflare and were often registered through NameSilo as file-share brands such as quickfly[.]online, cirrushare[.]co, and winsync[.]cloud. Others impersonated institutions, including tw-koryu[.]org, the misspellings heritiages[.]org and heritiage[.]org, and shinjirou[.]info. Some 2026 messages used a self-signed certificate that named a fictitious Kansas town, Millsstad, and a firm called Castro Inc. Others went out through residential proxies.
A February lure about military integration of Claude
July was not the first AI impersonation from this group. In February 2026 TA419 posed as a senior Anthropic employee and wrote an AI policy analyst at a United States think tank. The subject line was « Request for Feedback on Military Integration of Claude, » aimed at the fight over military use of Anthropic’s models. The shape matches the summer mail: a believable name, a live dispute, and no login page until someone replies.
Proofpoint has described a separate China-aligned actor, UNK_SweetSpecter, in earlier reporting on the SugarGh0st tool used against organizations that build generative AI. That work chased the labs. TA419’s 2026 mail chased the people who write the rules those labs live under, and Proofpoint also notes China-aligned interest in the semiconductor and rare-earth supply chains under the same models.
One suspicious feeling will not protect the next inbox
Engler noticed something slightly off, then checked with other people. That worked once, for someone who already knew Parker’s world. It does not travel to a junior analyst flattered by a committee that sounds official. The first email was clean on purpose, so many gateways had nothing to detonate, and the malicious step waited until a real-looking conversation had started.
Selena Larson of Proofpoint, speaking on 22 September 2026 with SiliconANGLE’s theCUBE, discussed how generative tools change the look of social-engineering mail. That segment is not about TA419. It is the wider problem around this incident. A note can be grammatical, specific, and still be a lure, so broken spelling is a weak defense for a policy shop.
The 1 October writeup is direct. « TA419 has consistently shown an interest in defense, national security, energy, international relations, and foreign policy targets, predominantly with a nexus to the US and Japan, » it says. « The targeting of AI policy experts represents an extension of that remit rather than a departure from it. » Proofpoint assesses that TA419 « will likely continue targeting think tanks and policy experts working on technologies, and in geographies, of particular interest to the Chinese government, » and that « these campaigns will likely also continue spoofing the identities of real subject-matter experts. »
Passkeys are the control that matches this phish
A one-time code did not stop the July pages. The code was typed into a relay of Microsoft’s own prompt, and the resulting session stayed with the operator. Proofpoint tells organizations in this target set to use phishing-resistant, origin-bound sign-in such as passkeys, which are bound to the real site and do not hand a password-and-code pair to a lookalike. It also tells individuals to treat a surprise note from a famous specialist as a pretext and to check it on a channel the sender did not pick.
A passkey will not rescue a thread already copied. What it changes is the next reply. The papers worth taking here lived in ordinary cloud mail, not a classified enclave, written by people who draft AI export-control language. Nouvelles du Hacker et Nouvelles internationales coverage has followed the high-volume version of the same theft, including a March 2026 Microsoft warning about impostor mail. TA419 is the boutique version: fewer than 10 inboxes, chosen for what those people were about to write. Crebo-Rediker’s biography is on the Council on Foreign Relations site, and more of this lane is filed under News at The AEGIS Alliance.
