- Island has identified 186 lookalike domains impersonating 27 AI brands, including ChatGPT, Claude, DeepSeek and Perplexity, in a technique dubbed TypoPrompting
- ClickFix attacks were discovered via a mistyped Claude address, and a mistyped Perplexity address led to a malicious browser extension network comprising 30 related extensions
- Island recommends organisations mitigate the risk by providing direct access to approved AI services, monitoring lookalike domains, and enforcing controls over browser extensions, application downloads and commands initiated through web sessions
Security researchers identify 186 lookalike domains across 27 AI brands, uncovering links to ClickFix attacks, malicious Microsoft Store-signed apps and browser hijacking.
MELBOURNE - Nov 16, 2026 - New research from Island, the agentic contol plane for enterprises, has revealed how simple typing mistakes when accessing popular AI tools can expose users to malware, fraudulent security checks and browser hijacking.
The research, TypoPrompting: From AI Typos to Malware, identified 186 lookalike domains impersonating 27 AI brands, including ChatGPT, Claude, DeepSeek, Perplexity and ElevenLabs. While some domains were parked or used for advertising redirects, researchers traced others to malicious activity designed to exploit users attempting to access legitimate AI services.
Key findings include:
- ClickFix attacks: A mistyped Claude address redirected users to a fake Cloudflare verification page, instructing them to execute a malicious PowerShell command. The underlying attack infrastructure was linked to 211 entry domains, including lookalikes of banks, insurers and enterprise software providers
- Microsoft Store-signed malware: Eight AI lookalike domains led to fake security checks distributing Net Guard, an application carrying a genuine Microsoft Store signature. Island's analysis identified browser-hijacking capabilities, including changing search settings and executing remotely supplied scripts
- Search-hijacking extensions: A mistyped Perplexity address led to a malicious search-redirecting browser extension, part of a network of 30 related extensions. As of 3 November, 17 remained listed in the Chrome Web Store, collectively displaying 972,000 declared users
Island has dubbed the technique TypoPrompting, describing how attackers exploit familiar AI brands to turn an attempt to access a legitimate tool into an instruction to verify, download or execute something malicious.
The findings highlight an emerging risk for organisations as employees increasingly rely on AI tools for everyday work. Island recommends providing direct access to approved AI services, monitoring lookalike domains and enforcing controls over browser extensions, application downloads and commands initiated through web sessions.
Read the full research, including technical findings and indicators of compromise: https://www.island.io/blog/typoprompting-from-ai-typos-to-malware
About us:
About Island
Island is the agentic control plane for enterprises, where organisations govern people and AI agents as they work across devices, browsers, applications, networks, and data. Island gives agents the identity management, access, guardrails, cost controls, and auditability they need to operate safely at scale. Millions of users trust Island to protect sensitive information, simplify enterprise work, and reduce reliance on legacy infrastructure. Island can be reached at [email protected].
Contact details:
Cath Arnott, Innovation Advisory
+61 432 405 155
