A mistyped character in an AI tool's web address can lead somewhere far more dangerous than an error page. In a report published on November 12, enterprise browser security company Island identified 186 domains that closely resemble the addresses of 27 well-known AI brands. The figure refers to web addresses, not confirmed infection cases. Some of the addresses redirect visitors to advertising pages, while others route them toward malicious sites.

A typo that lands somewhere else

Island has named the technique “TypoPrompting.” It is a modern variant of typosquatting, the long-standing practice of registering web addresses that differ from trusted domains by a letter or two. ChatGPT is the most frequently imitated brand, with 25 lookalike domains. Claude follows with 18, DeepSeek with 16 and Runway with 12, while Perplexity, Lovable and Character.ai have 11 apiece.

Not every lookalike address leads to the same destination. Some are built to generate advertising revenue. Others hand visitors to traffic brokers, which decide where to redirect them based on factors such as location, browser or device.

Three routes from mistyped name to malware

Island traced three malicious pathways tied to the lookalike addresses. The first began at cloude[.]ai, a misspelled version of Claude's domain. Visitors were redirected to a fraudulent Cloudflare verification page that asked Windows users to run a PowerShell command. That page spawned a network of 211 domains, including sites impersonating companies with no connection to the AI industry. Genuine checks of this kind are a routine fixture of the web, and PowerShell ships with Windows as a command-line administration tool.

The second pathway led to a fake “Security Check” promoting a tool called Net Guard. Although the application was available through the Microsoft Store, Microsoft's official marketplace for Windows software, Island found it could hijack browsers and execute remotely supplied scripts.

The third pathway involved 30 browser extensions that hijacked searches. Thirteen of them have been removed, leaving 17 still active with a combined 972,000 users, according to extension-store listings — a figure that reflects installs, not confirmed infections. Whether the remaining 17 get pulled from the stores remains to be seen.

Why ClickFix slips past automated defenses

The fake Cloudflare check relies on ClickFix. As Microsoft explains, the trick lies in convincing people to run malicious commands themselves inside familiar Windows tools, which can help attacks bypass automated defenses.

Browser-security firm Push Security cited Microsoft's 96.3% conversion figure for ClickFix malware delivery, and noted that ClickFix accounts for more than half of its monthly attack detections. Neither statistic measures Island's campaign.

Who profits when a visit goes wrong

Even a visit that stops short of infection can generate money. Island linked some redirects to advertising and affiliate payments through shared tracking IDs.

DNS-security company Infoblox found that more than 90% of visits in its parked-domain tests landed on scams, unwanted software or malware. Separately, software-supply-chain security firm Socket identified 319 search-hijacking extensions with roughly 13.4 million users. Those were independent investigations, not additional TypoPrompting victims.

What a hijacked browser could cost a business

For businesses, a hijacked browser can pose a far greater threat than an unwanted redirect. Attackers could gain access to corporate accounts or internal systems, although Island has not established any financial losses from this campaign. A browser holds a user's logged-in sessions to email, cloud services and internal tools, which is one reason browser security has emerged as a distinct enterprise product category.

Broader cybersecurity research suggests what may be at stake. Verizon's 2026 report found system intrusion to be the leading breach pattern at 61%, while IBM's 2026 research showed that AI-enabled malicious breaches cost an average of $6 million, above the overall breach average of $4.99 million. Both figures exclude any losses from Island's fake AI domains. As more companies adopt AI tools, the OECD's 2026 outlook flags digital security as a growing concern.

What readers can do now

The risks extend beyond mistyped addresses. In a separate campaign, Cryptopolitan previously reported that more than 260,000 Chrome users had installed at least 30 malicious extensions posing as AI helpers. Both cases show how attackers exploit people's trust in familiar AI brands.

The same precautions can help users avoid such traps: bookmark official AI websites, never run terminal commands prompted by a web page, and regularly review installed browser extensions. Enabling two-factor authentication adds another layer of protection, and businesses should restrict unauthorized add-ons.