A single mistyped letter in an AI website address can turn a routine visit into a malware trap. Researchers have linked lookalike domains for ChatGPT, Claude, and other tools to fake verification pages, suspicious installers, and search-hijacking extensions.
Browser security company Island identified 186 lookalike domains impersonating 27 AI brands in a Nov. 12 investigation. Some led to fake verification pages designed to trick users into running malicious commands, while others promoted suspicious applications or browser extensions that redirected searches.
For businesses, the danger goes beyond a mistyped URL. A convincing imitation of a trusted AI website could lead employees to execute malicious code or expose sensitive search queries. However, the 186 domains represent the researchers' inventory, not 186 confirmed infections.
AI lookalikes redirect users to different threats
In its TypoPrompting investigation, Island identified 25 lookalike domains for ChatGPT, 18 for Claude, and 16 for DeepSeek. The inventory also included parked domains and advertising redirects, so not every address was confirmed to deliver malware.
What a visitor sees can depend on their location, operating system, and browser. In one observed chain, the Claude lookalike cloude[.]ai passed through redirect and fingerprinting services before displaying a fake Cloudflare verification page.
The page instructed Windows users to open the Run dialog and paste a command copied to their clipboard. Island recovered a command designed to download a PowerShell script from a numerically encoded address and execute it in a hidden window. The researchers documented the delivery mechanism but did not report a confirmed infection count.
A separate ClickFix campaign involving malicious Custom GPTs also used fake verification instructions to get users to run commands. The two investigations concern different activity.
Three delivery routes put browsers and endpoints at risk
Island traced different paths from AI lookalike sites to unwanted actions:
Route | What the visitor encounters | What could follow |
ClickFix | A fake Cloudflare verification request | Execution of a downloaded PowerShell script |
Net Guard | A fake security check promoting a signed installer | Changes to browser settings, extensions, or remotely supplied scripts |
Search-hijacking extension | An extension installation page reached through a Perplexity typo | Search queries routed through a third-party provider |
The Net Guard route is particularly unusual. Eight AI-brand misspellings led to fake security checks offering an application marketed as a Wi-Fi and DNS safety scanner. Island found that the installer matched an existing Microsoft Store package, rather than a modified counterfeit.
"It is byte-for-byte identical to Net Guard as Microsoft distributes it," Island lead security researcher Oleg Zaytsev told The Hacker News.
Island's analysis found that the application could change a default search engine, alter browser profiles, install extensions, and execute JavaScript supplied by a remote server. A valid Store signature therefore did not establish that its behavior was safe.
Search hijacking can happen without permission warnings
A separate route began with preplexity[.]ai, a lookalike of Perplexity. The site led visitors to a Chrome extension that replaced the default search provider with a third-party endpoint, according to Island.
The extension did not need executable JavaScript or declared permissions to make that change. Workplace searches could consequently pass through an outside provider, potentially exposing internal hostnames, client names, or project details. Island linked 30 extensions to the same landing-page kit. The research did not establish how many people reached those extensions through the fake AI domains.
The technique differs from a previously reported Chrome extension compromise that pushed ClickFix malware. Both cases show why browser add-ons need oversight beyond checking their stated permissions.
How security teams can investigate and block these attacks
A visit to a misspelled domain does not by itself confirm that a computer was infected. Responders should determine whether the user executed a command, installed an application, or accepted an extension that changed browser settings.
Security teams can focus on three controls:
- Reduce risky visits: Give employees managed links to approved AI tools, and filter verified malicious lookalike domains rather than blocking unfamiliar domains indiscriminately.
- Watch for command execution: Correlate suspicious browsing with Windows Run or Terminal activity, hidden PowerShell processes, and unusual script downloads. Investigate promptly if someone followed a verification prompt.
- Audit apps and extensions: Restrict unapproved MSIX packages and browser add-ons, check unexpected search-provider changes, and evaluate signed software by its behavior as well as its certificate.
Google says Chrome Enhanced Protection can identify certain suspicious commands copied from webpages and warn users about potential ClickFix attacks. However, browser warnings are only one layer of defense and do not replace endpoint monitoring.
Island did not establish how many users were compromised through the identified domains. Still, the investigation highlights how attackers can turn familiar AI brands into convincing entry points for malicious commands, unwanted software, and search redirection.
For organizations, the lesson is straightforward: a trusted-looking AI website is not necessarily a trusted destination, and even a simple typing mistake deserves investigation when it leads to unexpected software or system changes.
Also read: Fake AI websites aren't the only threat to browser security. Learn how a separate AI browser extension scam exposed 260,000 users and what made those extensions dangerous.





