Fake AI domains funnel visitors into malware and search hijacks
Fake AI domains are being used to redirect people toward malware, dangerous commands and browser extensions that take over searches, according to cybersecurity researchers at Island. The campaign covers 186 bogus domains imitating 27 artificial intelligence brands, including Anthropic Claude, OpenAI ChatGPT, Google Gemini and Gemini Notebook, formerly NotebookLM, Microsoft Copilot, SpaceXAI Grok, DeepSeek, Mistral, Perplexity, Midjourney, ElevenLabs, Lovable and Cursor.
Island calls the delivery pattern TypoPrompting. It describes AI application lookalikes that turn an attempt to reach a familiar service into a request to verify, install or run something else. Examples include cloude[.]ai, cluade[.]ai, clsude[.]ai, chatlgpt[.]com, geimini[.]com, notebookklm[.]com, griok[.]com, preplexity[.]ai, deepceek[.]com, elevenlebs[.]io, copiot[.]com and midj0urney[.]org. Registration records show that some, including cloude[.]ai and cluade[.]ai, were registered as early as July 13, 2023.
The first route uses ClickFix, a social-engineering technique that persuades someone to run a command themselves. A parked domain, meaning a registered domain without a normal active website, starts a redirect chain that fingerprints the browser and collects the visitor’s public IP address before showing a fake Cloudflare verification page. On Windows, the page is designed to make the visitor run a command copied through the clipboard. On macOS, similar pages present terminal or security updates intended to make the user run a shell command. Another type of page offers an archive and tells the visitor to extract and run an executable.
Island said the fingerprinting script operates across a wider group of 211 entry domains. These also imitate banks, insurers and enterprise software, including Scotiabank, Charles Schwab, Blue Cross Blue Shield, Verizon, ServiceNow and Microsoft Azure. The purpose of checking the browser and public IP address is to decide what redirect or lure to show, rather than sending every visitor to the same page.
The use of parked domains for malicious redirection is not new. In December 2025, DNS threat intelligence company Infoblox described how domains without a current webpage can be monetised by sending visitors through traffic distribution systems, or TDSs. These systems are operated by third-party advertising platforms and can profile the host and device before delivering scams or malware. Infoblox said that when a domain owner enables direct search, traffic is sold to advertisers bidding on keywords and traffic characteristics, allowing a legitimate advertising model to be weaponised.
A second route uses eight typosquatted domains imitating Claude, DeepSeek, Copilot, ElevenLabs and Voice AI. Typosquatting means using a web address that differs slightly from a legitimate one, often by changing, adding or removing a letter. These domains lead to fake Security Check pages that deliver Net Guard, a browser-hijacking malware package.
Net Guard presents itself as a Wi-Fi and DNS safety scanner and has the same MSIX hash as a legitimate application with the same name. MSIX is a Windows application package format. However, Island found that the software can run attacker-controlled JavaScript, change the default search engine while the user is idle, edit browser profiles and install arbitrary Chrome extensions. Island said the download from the fake Security Check pages is unmodified Microsoft Store software, byte-for-byte identical to the version distributed by Microsoft and carrying Microsoft’s Store signature. The hidden browser-control capability is included in every copy and can switch on without asking, wait until the computer is not being used, take over Chrome’s search settings and run scripts inside the browser according to instructions from an operator server.
The third route centres on preplexity[.]ai, a lookalike of Perplexity. It redirects visitors to Quick Search, a browser extension available through the Chrome Web Store. An extension is a browser add-on that changes or adds functions. Island said the analysed extension needs neither executable JavaScript nor declared permissions for its main action. Its manifest, the configuration file that tells a browser what an extension does, changes the default search provider to find.flashysearches[.]com. Searches made through that provider are sent to a third party. In a workplace, those searches could contain internal hostnames, customer names and project details.
Quick Search belongs to a larger group of 16 extensions that includes GPT for Chrome, which has 80,000 users. Together, the extensions had 972,000 declared users as of November 3, 2026. The campaign therefore is not limited to people who mistype an AI service’s address once. A misleading domain can instead lead to software or an extension that changes what the browser does later, including when the user is not actively working.
For website owners and IT teams, the central risk is that a familiar-looking AI address can become the starting point for command execution, unwanted software installation or disclosure of workplace searches. AEU DNS provides private, secure DNS for everyone, which is relevant for organisations seeking a more controlled way to handle domain lookups.
Users should reach AI services through saved bookmarks or an organisation’s approved links, and should close any page that asks them to paste a command, run a terminal instruction or install a “verification” tool. They should also avoid installing browser extensions prompted by an unfamiliar website, review installed extensions regularly and remove anything they do not recognise. Businesses should remind staff that search queries can reveal internal names and project details, and should provide approved AI service links rather than relying on users to type addresses manually.