The Pulse

AI Domain Typos Led to Malware Lures, Island Research Finds

Island traced 186 lookalike domains across 27 AI brands, including addresses that led to fake verification pages and browser-hijacking software. One route delivered a Microsoft Store-signed package that researchers classified as a malicious browser hijacker.

AI.info Team · · fact-checked · against 1 source ↓ · Published 12 Nov 2026

Misspelled addresses for AI services led visitors to fake verification pages, browser-hijacking software and search-changing extensions in research covering 186 lookalike domains across 27 brands. The findings, published by enterprise browser company Island, include a Microsoft Store-signed app that researchers classified as a malicious browser hijacker.

The inventory included parked domains as well as malicious delivery routes. Island calls the pattern TypoPrompting: an AI-service lookalike leads a visitor toward a prompt to verify, install or run something else.

A typo leads to a Windows command

Among the addresses studied, ChatGPT had 25 lookalikes, Claude had 18 and DeepSeek had 16, according to the research. The set also included addresses resembling creative tools and company names.

In one traced route, a misspelling of Claude led through a traffic-distribution chain to a fake Cloudflare verification page. Island said the route collected information including the visitor’s country, operating system and browser before selecting a compatible lure.

Only selected visits received a lure, and other visits stopped earlier or received none. The Windows page instructed the visitor to open the Run dialog and paste a command that downloaded a PowerShell script and launched it in a hidden window.

The researchers connected the gate to 550 matching scan records across 211 entry domains between April 2025 and November 2026, after excluding decoy redirects. The wider set included lookalikes of banks, insurers and enterprise software, with separate lures for Windows and macOS and pages offering archive downloads.

Store-signed package examined

A second route took visitors from eight typos of Claude, DeepSeek, Copilot, ElevenLabs and Voice AI to fake security-check pages, according to Island. Those pages delivered Net Guard, an app presented as a Wi-Fi and DNS safety scanner.

The downloaded package matched the hash of the Net Guard Microsoft Store listing and carried Microsoft’s Marketplace signature, the researchers found. Their analysis said the app could change browser search settings, run remotely supplied JavaScript through DevTools and alter browser profiles or install extensions.

Those features were enabled by default in the code analyzed, Island said. Its researchers also examined Privacy Keeper and MyFlixly Browser packages, classified all three as malicious browser hijackers and said they shared the same underlying code.

The AI typos formed part of a larger installer funnel spanning 191 related landing domains, about 45 product names and 1,437 source domains, the research found. Those source domains included lookalikes involving banks, insurers and telecoms as well as AI tools.

Extensions redirected searches

A separate Perplexity typo led through a related-search page to an installation page for a Chrome Web Store extension that changed the default search engine. Island said its manifest routed address-bar searches through a third-party search endpoint, despite the listing promising Bing search.

The team linked the extension to a kit behind 30 extensions and said every listing it checked was unlisted. Island said 13 had since been removed; the 17 still listed displayed 972,000 users in total on November 3.