Proofpoint flags hidden AI command injections in emails and fake Windows app sites delivering malware

Security researchers at Proofpoint have identified a new technique called indirect prompt injection, where attackers embed invisible commands in ordinary emails, documents, calendar invites and online ads. The hidden “white‑on‑white” text is designed to be read by AI mail‑filtering agents, potentially causing them to execute unsafe actions or expose data. Although the tools are still experimental and no large‑scale abuse has been seen, the researchers warn that the method could add a new layer of risk to existing phishing attacks.

Separately, security analysts have uncovered a campaign of look‑alike websites that impersonate popular Windows utilities such as PowerToys, EasyBCD and CrystalDiskMark. The sites initially present legitimate‑looking download pages, then later replace the links with malware that can install remote‑access tools or other unwanted software. The activity was first reported by developer Bogdan X and further tracked by Check Point Software Technologies, which noted that the infrastructure can harvest traffic before swapping in malicious payloads. Infections have already been confirmed on several impersonated apps, highlighting a growing threat to Windows users.