Severity: High | Priority: 1
Island has identified a campaign it calls TypoPrompting. It uses 186 lookalike domains impersonating 27 popular AI services, including Claude, ChatGPT, Gemini, Copilot, Grok, DeepSeek, Perplexity, Midjourney and ElevenLabs. The domains catch users who mistype an AI service’s address, or follow a misspelled link, and send them through traffic-distribution and fingerprinting services. Those services decide what each visitor sees. Users don’t have to visit a suspicious site; a single typo in an address they use every day is enough.
Visitors end up on one of three main paths. On the first, Windows users see a fake Cloudflare verification page that asks them to paste a command into the Run dialog. The command quietly downloads a PowerShell script from an IP address written as a decimal number to look less suspicious, then runs it hidden with script restrictions bypassed. On the second, visitors see fake “Security Check” pages that push Net Guard and related apps. These are browser-hijacking programs distributed through the Microsoft Store that can change search settings, run remote code in the browser and install extensions. On the third, visitors are steered into installing search-hijacking Chrome extensions; Island links the kit to extensions with about 972,000 declared users. A few of the domains have also been used in the past to deliver information stealers and Android banking malware.
Give staff direct links or managed browser bookmarks to the AI services your organisation has approved, instead of relying on typed addresses. Block or monitor lookalikes of those services at DNS level. Remind users that a real verification page never asks them to paste or run a command.
On endpoints in your network, look for the following:
Remove anything you find. If the PowerShell script ran, treat the machine as compromised.