Cybersecurity researchers have uncovered a campaign involving 186 fraudulent websites impersonating popular artificial intelligence platforms, including ChatGPT, Claude, Gemini and Microsoft Copilot. The fake domains are being used to expose unsuspecting visitors to malware, deceptive verification pages and malicious browser extensions.
The campaign, identified as “TypoPrompting” by cybersecurity firm Island, exploits typing errors made by users attempting to access legitimate AI services. Instead of reaching the platforms they intend to visit, users who enter slightly misspelled web addresses may be redirected to malicious websites designed to trick them into downloading or executing harmful software.
According to Island, the operation covers 186 fraudulent domains impersonating 27 AI brands. The targeted services include Anthropic Claude, OpenAI ChatGPT, Google Gemini, Gemini Notebook, Microsoft Copilot, Grok, DeepSeek, Mistral, Perplexity, Midjourney, ElevenLabs, Lovable and Cursor.
Researchers identified three major methods through which the fraudulent domains expose visitors to security threats.
The first involves ClickFix-style attacks, in which victims are presented with fake Cloudflare verification pages designed to persuade them to copy and execute malicious commands on their computers.
The second method uses deceptive security-check pages to distribute a browser-hijacking malware known as Net Guard. The third redirects visitors to malicious browser extensions capable of manipulating search activity and redirecting users’ searches through third-party services.
Some of the fraudulent web addresses identified 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.
An examination of domain registration records revealed that some of the suspicious addresses, including cloude[.]ai and cluade[.]ai, were registered as far back as July 13, 2023.
Island explained that the ClickFix attack begins when a visitor lands on a parked domain, which subsequently initiates a series of redirects. During the process, scripts collect information about the visitor’s browser and public IP address before presenting a fake verification page.
The researchers also discovered that the fingerprinting infrastructure extends beyond AI platforms. According to the report, the same script covers 211 entry domains impersonating organisations in the banking, insurance and enterprise software sectors.
These include lookalike websites associated with Scotiabank, Charles Schwab, Blue Cross Blue Shield, Verizon, ServiceNow and Microsoft Azure.
The fraudulent pages use different techniques depending on the operating system and the type of deception being attempted.
On Windows computers, victims may encounter fake verification pages that instruct them to execute commands stored in their clipboard. Mac users may instead be presented with supposed terminal or security updates that require running shell commands. Other pages offer archive downloads and instruct visitors to extract and launch executable files.
Researchers also identified eight typosquatted domains impersonating services such as Claude, DeepSeek, Copilot, ElevenLabs and Voice AI. These websites redirect visitors to fake security-check pages that eventually deliver Net Guard.
Although Net Guard presents itself as a Wi-Fi and DNS security scanner, Island found that the application contains capabilities that could be abused by attackers.
The software can execute attacker-controlled JavaScript, modify browser profiles, change a user’s default search engine while the computer is idle and install unauthorised Chrome extensions.
The researchers noted that the software shares an MSIX hash with a legitimate application bearing the same name, making the malicious version particularly concerning for unsuspecting users.
The third attack pathway involves fraudulent websites impersonating Perplexity through the domain preplexity[.]ai. Visitors are redirected to a Chrome Web Store extension called Quick Search.
According to Island, the extension does not require executable JavaScript or declared permissions to carry out its primary activity. Instead, its configuration changes the browser’s default search provider to find.flashysearches[.]com, meaning searches conducted through the browser are routed to a third party.
The researchers warned that this could create serious privacy concerns, particularly in workplaces where search queries may contain sensitive information such as internal computer names, customer details and confidential project information.
Quick Search is reportedly part of a wider network of 16 browser extensions, including GPT for Chrome, which has approximately 80,000 users.
Collectively, the extensions had declared 972,000 users as of November 3, 2026.
Island warned that the growing reliance on AI platforms is creating new opportunities for cybercriminals to exploit users through deceptive web addresses.
The company advised organisations and individuals to avoid relying on manually typed addresses when accessing important online services. Instead, users should access AI platforms through verified bookmarks or trusted links and carefully inspect website addresses before entering sensitive information or downloading software.
The findings highlight how a seemingly minor typing error can expose users to malware, browser manipulation and potential data theft.