Chinese spies impersonate White House, Anthropic figures to phish AI policy experts

A China-aligned group known as TA419 conducted credential phishing campaigns targeting AI policy experts in the United States in July 2026. The group impersonated individuals including a former White House official and an economist, among others, to gain access to cloud accounts of these experts. Specific impersonations included a former member of the White House Office of Science and Technology Policy leadership team and an economist and foreign policy expert. Additionally, the group impersonated a senior Anthropic employee in February 2026 to target an AI policy analyst at a US think tank, using email subjects referencing debates over Claude models.
The phishing attacks began by inviting targets to join fictitious groups or contribute to reports on AI export controls and supply chains, initially without malicious links. Upon response, the attackers provided shortened URLs leading to fake OneDrive credential phishing pages. These pages employed layered techniques, routing traffic through group-controlled domains using Cloudflare's CDN and a fake loading screen. The final phishing page utilized a browser-in-the-browser tool to capture session cookies upon simulated login, allowing attackers to obtain cloud account access credentials, including Multi-Factor Authentication codes.
The activity is linked to broader Chinese intelligence objectives related to understanding US AI policy and regulatory developments amidst strategic competition and export control discussions involving the US and China.

* A China-aligned group tracked as TA419 conducted credential phishing campaigns in July 2026.
* The group impersonated a former member of the White House Office of Science and Technology Policy leadership team.
* The group impersonated an economist and foreign policy expert.
* In February 2026, the group impersonated a senior Anthropic employee.
* One phishing email had the subject line “Request for Feedback on Military Integration of Claude.”
* Initial emails invited targets to join fictitious groups or contribute to AI supply chain reports.
* Replies led to shortened URLs pointing to fake OneDrive credential phishing pages.
* The phishing process involved traffic routing through group-controlled domains using Cloudflare CDN and a fake loading screen.
* The final page used a browser-in-the-browser tool to capture session cookies after fake login, enabling credential theft.
* The group registered domains impersonating organizations such as heritiages[.]org for the Heritage Foundation.

The documented methodology reveals a sophisticated blending of social engineering and technical infrastructure aimed at intelligence gathering within a high-stakes geopolitical context. The attack leverages the public discourse surrounding AI policy—specifically export controls and military integration debates involving entities like Anthropic—as a vector for building initial trust, moving from benign requests to credential harvesting. This reflects a pattern where abstract strategic competition is translated into tangible, technically sophisticated operational objectives targeting specific knowledge silos.
The use of cloud service mimicry (OneDrive) combined with advanced browser simulation techniques suggests an adaptation of established phishing tactics into a persistent, multi-stage infiltration method designed for maximum session data extraction rather than simple initial login capture. The infrastructure setup—utilizing CDN obscuration and domain impersonation linked to geopolitical entities—demonstrates an understanding of operational security necessary for state-level intelligence operations. This elevates the concern beyond typical cybercrime to targeted influence campaigns seeking specific regulatory insights.
The implication is that the current friction points in US-China AI strategy are being actively exploited through digital deception to map out the internal workings and policy consensus among key decision-makers. This shifts the focus from abstract competition to the concrete vulnerability of information flow itself. If these methods are scalable, it suggests that regulatory uncertainty, rather than purely technical gaps, becomes a primary domain of strategic contestation, forcing individuals in high-level policy circles into continuous vigilance regarding digital identity and contextual understanding.
Bridge Questions:
What other specific aspects of US AI policy, beyond export controls, might be targeted for intelligence gathering through similar digital means? How does the normalization of state-aligned infrastructure influence public and private sector responses to AI governance? What mechanisms exist outside of traditional cybersecurity frameworks to evaluate the threat posed by information exploitation in strategic competition?