Artificial Intelligence · Cybersecurity
Phishing pages impersonated AI ad tools to steal logins
Island researchers said a phishing platform copied advertising tools for AI chatbots and used fake sign-in windows to collect passwords, MFA codes and device data through browser-in-browser pages.
- Island researchers Oleg Zaytsev and Ofek Ronen said the platform copied ad products tied to Gemini, Claude, ChatGPT, Perplexity, Meta Muse and Manus.
- The fake pages used a Connect prompt and browser-in-browser windows to collect passwords and MFA codes.
- The researchers said the system fingerprinted devices, sent data to /api/send/ip over Socket.IO, and let an operator choose the next login step.
- One page, museads.ai, appeared on 16 September 2026 and claimed to be an AI ads manager.
Island said the pages impersonated advertising products linked to AI chatbots, including Gemini, Claude, ChatGPT, Perplexity, Meta Muse and Manus. The designs were built around a Connect action, and the fake window showed a trusted-looking address while the real browser stayed on the phishing site.
On museads.ai, a Prompt Box and Connect button triggered the browser-in-browser trick to capture Google, Meta, TikTok and Okta credentials. The victim’s device was fingerprinted, the data was sent to /api/send/ip over Socket.IO, and operator commands were exchanged with the page in real time.
Zaytsev said the same parser handled two Socket.IO event names, operator-command and telegram-command. He said the operator could ask for another password, show SMS or authenticator challenges, present Google prompt screens, or switch between Okta login steps.
Island said the wider platform also included Google Ads refund and payment pages and recruitment-themed sites for Tesla, Louis Vuitton, Nike and Adecco. It said earlier source code for older versions had been exposed in public GitHub repositories.
The company said the AI ads campaign targeted agency staff, media buyers and manager-account administrators, and was likely meant to monetise ad accounts or sell them. It also said phishing-resistant authentication, review of ad-control changes and checks on AI integrations could reduce the risk.
Named in this story
People
- Oleg Zaytsev
- co-authored Island’s report and explained the page logic
- Ofek Ronen
- co-authored Island’s report
Companies
- Island
- reported and analysed the phishing platform
- had its sign-in pages and ads products mimicked
- Meta
- launched Muse and had its services mimicked
- Anthropic
- had Claude imitated
- OpenAI
- had ChatGPT imitated
- TikTok
- had its workflows targeted in the fake sign-in flow
- Okta
- had its tenant and login screens imitated
- Mimecast
- published a July 2026 report on ad account theft
- Tesla
- was named in recruitment-related phishing sites
Products and systems
- Google Gemini
- was impersonated by a fake ad portal
- Anthropic Claude
- was impersonated by a fake ad portal
- OpenAI ChatGPT
- was impersonated by a fake ad portal
- Perplexity
- was impersonated by a fake ad portal
- Meta Muse
- was impersonated by a fake ad portal
- Manus
- was impersonated by a fake ad portal
- Socket.IO
- carried device data and operator commands
- Next.js
- was part of the shared site stack
- VietCredCare
- was named by Mimecast as a malware family tied to ad account theft
- DuckTail
- was named by Mimecast as a malware family tied to ad account theft
- NodeStealer
- was named by Mimecast as a malware family tied to ad account theft
- PXA Stealer
- was named by Mimecast as a malware family tied to ad account theft
Places
- museads.ai
- hosted one of the phishing pages
How the source tells it
Mostly neutral technical reporting; any menace came from the subject matter and the researchers’ own descriptions rather than editorial flourish.
No emotive framing was found in the original.