Fake AI Sites: 17 Trapped Extensions Still Listed in Early November

Domains with names similar to AI services are directing internet users towards various traps, according to security researchers at Island. Their investigation describes browser hijacking, as well as malicious commands executed under the guise of verification.
The Chrome Web Store Still Hosted 17 Detected Extensions
As of November 3, 2026, 17 extensions deemed malicious by Island remained available on the Chrome Web Store. Researchers had identified 30; 13 had been removed. These extensions replace the default search engine used in the browser, explains Island. Attackers can then profit from monetized searches or resell the collected queries.
A removal of the remaining extensions was anticipated in the following days or weeks, though there was no confirmation. The status of the sites themselves remains uncertain: Island reports no domain seizures, and some may still be operational. Security researchers typically make their findings public after helping to reduce risk, notably through seizures or takedowns, but such intervention is not documented here.
Island Links 211 Domains to a Common Infrastructure
By studying a shared infrastructure, Island associated 211 entry domains with this operation. The imitations are not limited to AI: banks, insurers, telecom operators, and enterprise software publishers are also among the identities being impersonated. Researchers estimate that the scope likely exceeds just AI assistants and tools. Multiple groups may be employing the same method, rather than a single operator.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
For AI brands alone, Island identified 186 deceptive domains targeting 27 brands, including Midjourney and Cursor. The attackers reportedly purchased these addresses by making slight alterations to the names: chtgpt.com mimics ChatGPT, while cluade.ai evokes Claude. It is this domain squatting campaign that researchers uncovered.
ChatGPT tops the list with 25 imitations, followed by Claude with 18, DeepSeek with 16, and Runway with 12. Perplexity, Lovable, and Character.ai each have 11 deceptive addresses.
Net Guard Can Modify Browser Settings
According to Island, the displayed destination after a visit depends on the country, operating system, and browser. Internet users may encounter a fake Cloudflare control, a supposed security verification, or an invitation to install extensions. These pathways serve to distribute malicious software or extensions, or to trigger a ClickFix attack.
The alleged security verification notably offers Net Guard, an application that appears legitimate. Island indicates that it is signed and distributed in a package corresponding to its official listing on the Microsoft Store. However, it can hijack the browser, replace its default search engines, modify profiles, add extensions, and execute JavaScript remotely.
The Cloudflare lure relies on ClickFix: under Windows, the victim is prompted to copy and then execute malicious PowerShell commands. The scripts retrieved and executed come from systems controlled by the attackers. However, Island does not identify the final payload or the actions it performs.
Recommended precautions include carefully checking the entered address before opening a page. Saving regularly used tools in bookmarks is another option.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.