Island: Finds 186 spoof AI brand domains and tracks three attack paths
Security company Island released research on November 12, 2026, written by Lead Security Researcher Oleg Zaytsev, tracking 186 domains spoofing 27 well-known AI brands. ChatGPT was spoofed the most, with 25 similar domains; Claude had 18, DeepSeek had 16, and Runway had 12. Some domains redirected visitors to ad pages, while others pointed to malicious websites.
The research describes the attack method as TypoPrompting, which exploits spelling mistakes when users enter AI tool URLs. Island tracked three paths: first, visitors, such as those coming from cloude[.]ai, were redirected to a fake Cloudflare verification page that induced Windows users to run PowerShell commands, with 211 entry domains related to the gateway service; second, a page disguised as a "security check" promoted the Net Guard app (signed by Microsoft Store), which could hijack browsers and run remote scripts; third, 30 browser extensions involved in search hijacking, of which 13 have been removed from the store, while the remaining 17 have a cumulative claimed user base of 972,000.
Island said the 186 fake domains are a count, not confirmed infection cases, and it has not yet confirmed that the activity caused any losses. Registration records for some domains can be traced back to July 2023, and the traffic distribution system decides the final destination based on the visitor's location, operating system, and browser.
Original source:
Title: TypoPrompting: From AI Typos to Malware
Publisher: Island (Oleg Zaytsev, Lead Security Researcher)
Publication date: 2026-11-12
URL: island.io/blog/typoprompting-from-ai-typos-to-malware
Source type: official research
Verification: The main body content was actually read for this report, verification date 2026-11-12
Supporting content: 186 spoofed domains, 27 brands and specific counts, the name TypoPrompting, details of the three attack paths (including PowerShell, Net Guard, and extension user counts), the statement that 186 is a count rather than confirmed infections, and the registration time of some domains.
ABAB AI Insight
Island was founded in 2020 by Dan Amiga (with an Israel Unit 8200 background) and Mike Fey (former Symantec president and McAfee CTO), and focuses on enterprise browser security. The company completed a $250 million Series E financing at an estimated valuation of about $4.8 billion in March 2025, and completed a $400 million Series F financing at a valuation of $6.4 billion in September 2026, bringing cumulative financing to more than $1 billion. Oleg Zaytsev is the company's Lead Security Researcher; he previously worked on threat hunting at Guard.io, and this research directly reflects his threat-tracking capability.
This research is not a capital transaction, but a threat-intelligence output related to the product. Through its enterprise browser, Island intercepts browser-to-shell handoffs, clipboard commands, and similar actions, turning this type of TypoPrompting attack into a product defense point. Funding came from Sequoia, Coatue, Insight Partners, and others, and is used to expand the platform into AI agent governance and data protection, rather than directly for domain monitoring itself.
Comparable cases include parking-domain traffic recorded by Infoblox in December 2025 that, after analysis, entered a malicious redirection chain, and midj0urney[.]org in 2023, which once distributed Aurora stealer. The similarity lies in the use of spelling mistakes and traffic distribution pipelines; the difference is that this case uses AI brands as the main entry point and includes a browser-hijacking app signed by Microsoft Store, as well as 211 entry domains spanning non-AI areas such as banking and insurance.
The concrete change is that after AI tools became everyday entry points, spelling mistakes no longer merely return an error page, but can directly lead into distribution chains for ClickFix, Store-signed malicious apps, or search-hijacking extensions. Enterprise users can block such flows through browser-layer controls such as Island, while individual users still rely on manually checking domains; the indicator to watch is whether confirmed infections or loss reports targeting specific AI brands appear later.
ABAB News · Cognitive Law
After AI tools become everyday entry points, spelling mistakes change from failed redirects into entry points for attack chains.
Enterprise browser companies use threat research to demonstrate product value while accumulating attack intelligence.
An inventory of spoofed domains does not equal confirmed infections, but it has already formed a scalable distribution pipeline.