Fake AI-tool lures tied to 1,258 malware analyses

ANY.RUN tied 1,258 analyses to fake Claude, ChatGPT, and other AI lures delivering infostealers; sources disagree on Claude’s share.

ANY.RUN linked 1,258 public sandbox analyses from April 15 to October 5, 2026, to campaigns impersonating Claude, ChatGPT, Codex, GitHub Copilot, Gemini, and other AI brands. It reported Claude at 601 analyses (48%) and ChatGPT or Codex at 418 (33%), together 81%, while Dark Reading puts Claude at 60% and ChatGPT or Codex at 33% and describes only a six-month span; The Hacker News dates the window as April to October 2026. Lures on GitHub, GitLab, Cloudflare Pages, Google Sites, sponsored search, shared chats, SEO-poisoned results, malicious ads, and counterfeit sites used pasted commands—at least 45%, or nearly half—along with mshta abuse, trojanized installers, and fake sign-ins. Reported results include infostealers, remote-access trojans, browser-session theft, and access to developer, cloud, advertising, and billing accounts, with Dark Reading also citing exposure of source code and cloud and AI API keys. Sources disagree on US exposure: ANY.RUN ranks US submissions second after India and first in major Claude and ChatGPT campaigns, while The Hacker News says US businesses were 16% of country-attributed targets.

• ANY.RUN linked 1,258 public sandbox analyses from April 15 to October 5, 2026, to fake-AI campaigns; The Hacker News dates the window April–October 2026, and Dark Reading says six months.
• Impersonated brands include Claude, ChatGPT, Codex, GitHub Copilot, Gemini, and other AI tools.
• ANY.RUN counted Claude at 601 analyses (48%) and ChatGPT or Codex at 418 (33%), or 81% combined; Dark Reading instead says Claude was 60% and ChatGPT or Codex 33%.
• At least 45% of lures told victims to paste a command; Dark Reading calls ClickFix-style prompts nearly half. Other methods included trojanized installers and fake sign-in windows.
• Lures appeared on GitHub, GitLab, Cloudflare Pages, Google Sites, sponsored search, SEO poisoning, malicious ads, shared chats, and counterfeit download or login sites.
• Reported outcomes include infostealers, remote-access trojans, stolen browser sessions, and access to developer, cloud, advertising, and billing accounts; Dark Reading adds exposure of source code and cloud and AI API keys.
• US exposure differs by source: ANY.RUN ranks US submissions second after India and first in major Claude and ChatGPT campaigns, while The Hacker News says US businesses were 16% of country-attributed targets.