MediumEcosystem changeOngoing

TypoPrompting: 186 fake AI domains imitate ChatGPT, Claude and more

Island found 186 lookalike domains of 27 AI brands leading to ClickFix lures and hijackers; bookmark AI tools and never paste commands.

Published
Updated
Exploitation
Yes, Island reports live lookalike domains delivering ClickFix lures, Net Guard and search-hijacking extensions, as covered by three outlets

What happened

Researchers at the security company Island named the pattern TypoPrompting. They describe it as AI-app lookalikes that turn an attempt to reach a familiar tool into a prompt to verify, install or run something else, according to The Hacker News. The campaign spans 186 domains across 27 AI brands, including Claude, ChatGPT, Gemini, Copilot, Grok, DeepSeek, Mistral, Perplexity, Midjourney, ElevenLabs, Lovable and Cursor (The Hacker News). Cyber Security News reports the research was published on November 12, 2026, by Island lead security researcher Oleg Zaytsev.

ChatGPT has the most lookalikes at 25, followed by Claude with 18 and DeepSeek with 16, per Cyber Security News and GBHackers. Examples include cloude[.]ai, cluade[.]ai, clsude[.]ai, chatlgpt[.]com, preplexity[.]ai and midj0urney[.]org (The Hacker News). The Hacker News says some, such as cloude[.]ai and cluade[.]ai, were registered as early as July 13, 2023.

A parked page hands the visitor to a traffic distribution system. It picks a destination from the visitor’s location, operating system and browser, so one person sees a harmless page and another gets a lure (GBHackers, Cyber Security News). The sources describe three routes:

  • ClickFix lure. The redirect chain fingerprints the browser and collects the public IP address before showing a fake Cloudflare verification page (The Hacker News). GBHackers reports the page tells the victim to paste a clipboard command into the Windows Run dialog. The command uses curl.exe to fetch W901.ps1 from 3112851459, which is 185.138.88[.]3 in decimal form, and runs it in a hidden PowerShell window with an execution-policy bypass. The decimal form leaves no dotted IP address for a pattern-based filter to match. The same gate serves 211 entry domains, some imitating Scotiabank, Charles Schwab, Verizon and Microsoft Azure. macOS visitors get a fake Terminal update that runs a shell command, and others get an archive to extract and run (The Hacker News, GBHackers).
  • Net Guard. Eight typosquats of Claude, DeepSeek, Copilot, ElevenLabs and Voice AI lead to fake “Security Check” pages that deliver Net Guard (The Hacker News). Zaytsev told The Hacker News the download is the unmodified Microsoft Store app, byte-for-byte identical and carrying Microsoft’s Store signature. Island says it can run attacker-controlled JavaScript, change the default search engine while the user is idle, edit browser profiles and install Chrome extensions. Cyber Security News and GBHackers report that Privacy Keeper and MyFlixly Browser share the same code and are classed as browser hijackers.
  • Search-hijack extensions. The typo preplexity[.]ai leads to a Chrome extension, Quick Search, whose manifest sets find.flashysearches[.]com as the search provider. It needs no declared permissions or executable JavaScript (The Hacker News). Island notes that searches in a workplace can include internal hostnames, customer names and project details.

The outlets differ on the extension counts. The Hacker News reports 16 live Chrome Web Store listings, including GPT for Chrome with 80,000 users, and 972,000 declared users as of November 3, 2026. GBHackers and Cyber Security News say Quick Search is one of 30 extensions from a shared kit, with 13 removed and 17 still listed showing 972,000 users on November 3. GBHackers also refers to 16 live listings in an image caption. The exact live count is not confirmed.

Some typos have delivered payloads directly, per GBHackers. In October and November 2026, chatlgpt[.]com offered Installer_ChatGPT.zip, which 9 of 16 MetaDefender engines flagged. In 2023, midj0urney[.]org served Aurora stealer, and in early 2025 deepsek[.]icu served an Android package.

Am I affected?

You may be exposed if anyone in your organization reaches AI tools by typing addresses rather than using bookmarks, or if Windows or macOS users can run pasted commands. The sources do not give a count of victims, and they do not confirm whether the domains have been taken down.

Things worth checking, based on the indicators the sources list:

  • Browser or proxy logs for visits to the lookalike domains named above, and to the redirect and gate hosts reported by Cyber Security News (for example securetrackingnode[.]com, biomimicrydevtech[.]buzz, blockchainfuturedev[.]buzz and deezyj[.]com).
  • Hidden PowerShell windows, execution-policy bypass flags, curl.exe use and decimal-format IP addresses on Windows endpoints (GBHackers).
  • Installed Net Guard, Privacy Keeper or MyFlixly Browser packages, and Chrome extensions that set find.flashysearches[.]com as the search provider.
  • Unexpected changes to the default search engine in Chrome.

What to do now

  1. Give staff bookmarks or managed links to the AI tools you sanction, and monitor for lookalike domains of those tools. This is Island’s recommendation as reported by GBHackers and Cyber Security News.
  2. Tell users that a real verification page never asks them to paste a command into the Run dialog or a terminal (Cyber Security News).
  3. Correlate browsing activity with Run-dialog or Terminal use, and flag hidden PowerShell, bypass flags and decimal IP addresses (GBHackers).
  4. Judge MSIX files by package family, publisher and hash, not by the Store signature, because a Store signature does not describe browser behavior (GBHackers).
  5. Enforce central search-provider and extension policy in managed browsers (GBHackers, Cyber Security News).
  6. Hunt the whole chain, from entry domain through redirects to the endpoint change (GBHackers, Cyber Security News). Cyber Security News publishes SHA-256 hashes, package IDs and extension IDs that can be used for matching.

Update, 17 November 2026

  • TechRadar reports that Island lists Runway with 12 typosquatted domains after ChatGPT (25), Claude (18) and DeepSeek (16), and Perplexity, Lovable and Character.ai with 11 each. (techradar.com)
  • TechRadar says Island does not state what the final payload of the ClickFix PowerShell chain is or what it does. (techradar.com)
  • TechRadar says the campaign was likely run by numerous actors using the same technique rather than a single actor, and that researchers warn it remains active. (techradar.com)
  • Island’s Oleg Zaytsev says the hidden browser control in Net Guard comes with every copy, whether installed from the Microsoft Store or from a typo site. He says it switches itself on without asking, waits until the user stops using the computer, then takes over Chrome’s search settings and can run scripts in the browser on instructions from the operator’s server. (thehackernews.com)
  • The Hacker News says Net Guard presents itself as a Wi-Fi and DNS safety scanner and matches the MSIX hash of the legitimate app of the same name. (thehackernews.com)

Sources