Security company Island has documented a campaign it calls TypoPrompting: 186 lookalike domains imitating 27 AI brands, including Claude, ChatGPT, Gemini, DeepSeek and Perplexity. The Hacker News reported the findings on November 12.
Ordinary typing mistakes can lead to three different payloads. One route presents a fake Cloudflare check and a ClickFix lure that runs a hidden PowerShell command. Another serves Net Guard, a Microsoft Store-signed app containing browser-hijacking malware. A third promotes search-hijacking browser extensions.
Island says 17 extensions built from the same kit were still listed when it published its research, together declaring 972,000 users; 13 others had been removed. The lookalikes do not mean the genuine AI services were breached. The risk begins before a prompt is entered, when a misspelled address sends a user to an attacker-controlled page. Users should check the exact domain and avoid running commands or installing software requested by an unexpected verification page.