Fake AI Websites Trick Users Into Running Malicious Commands and Installing Hijacking Extensions
Researchers at security company Island have identified a campaign they call TypoPrompting, in which lookalike domains imitating popular AI tools are used to deceive visitors. The campaign spans 186 fake domains covering 27 AI brands, including Anthropic Claude, OpenAI ChatGPT, Google Gemini, Microsoft Copilot, DeepSeek, Mistral, Perplexity, Midjourney, ElevenLabs, Lovable and Cursor. Some domains, such as cloude[.]ai and cluade[.]ai, were registered as early as July 13, 2023.
The campaign sends visitors down three routes. One leads to a ClickFix-style lure, where a fake Cloudflare verification page serves a PowerShell command. In that route, a redirect chain first fingerprints the browser and collects the visitor's public IP address. A second route leads to "Security Check" pages that deliver a browser-hijacking malware called Net Guard. The third sends users to a family of search-hijacking extensions, with 16 live Chrome Web Store listings and 972,000 declared users.
The tactic builds on an older problem. In December 2025, Infoblox described how parked domains, which have no real web page, can be monetised by redirecting visitors through advertising traffic networks to scams and malware. Island's researchers say these lookalikes turn an attempt to reach a familiar tool into a prompt to verify, install or run something else.