186 lookalike AI domains turn ChatGPT and Claude typos into malware

Island, an enterprise-browser security company, published the findings on November 12: 186 web addresses that imitate the domains of 27 known AI brands. Island calls the technique TypoPrompting. The count refers to addresses, not confirmed infections. ChatGPT is the most impersonated brand with 25 lookalike domains, followed by Claude at 18, DeepSeek at 16 and Runway at 12. Perplexity, Lovable and Character.ai each have 11.

The fake addresses don't all lead to the same place. Some just harvest advertising revenue. Others hand visitors to traffic brokers, which pick a destination based on the person's location, browser or device.

Island traced three malicious routes. The first starts at cloude[.]ai, a misspelling of Claude, and dumps visitors on a fake Cloudflare verification page that tells Windows users to run a PowerShell command; that page is tied to a network of 211 domains, some impersonating companies with nothing to do with AI. The second pushes a bogus Security Check for an app called Net Guard, which Island says is downloadable from the Microsoft Store but can hijack browsers and execute remotely supplied scripts. The third covers 30 search-hijacking browser extensions. Thirteen have been removed; the remaining 17 sit in stores with a combined 972,000 users, a figure pulled from extension listings rather than confirmed infections.

The fake Cloudflare page leans on ClickFix, which Microsoft describes as tricking people into running malicious commands inside familiar Windows tools so attacks slip past automated defenses. Push Security cited Microsoft's 96.3% ClickFix malware-delivery conversion figure and said ClickFix accounts for more than half of its monthly attack detections; neither number measures Island's campaign. Even a visit that stops short of infection pays: Island linked some redirects to ad and affiliate payments through shared tracking IDs. Infoblox found that over 90% of visits in its parked-domain tests reached scams, unwanted software or malware, while Socket separately identified 319 search-hijacking extensions with roughly 13.4 million users.

For companies, the payoff is bigger than a stray redirect. A hijacked browser can open the door to corporate accounts and internal systems, though Island has not established any financial losses from the campaign. Verizon's 2026 report puts system intrusion as the leading breach pattern at 61%, and IBM's 2026 research prices AI-enabled malicious breaches at an average of $6 million against an overall breach average of $4.99 million. Both exclude Island's domains. The OECD's 2026 outlook flags digital security as a rising concern as AI adoption spreads.

Cryptopolitan has reported a similar pattern before: more than 260,000 Chrome users installed at least 30 malicious extensions posing as AI helpers. The defenses are plain — bookmark official AI sites, never run terminal commands a web page tells you to run, and audit installed extensions. Two-factor authentication adds another layer, and businesses should block unauthorized add-ons.