Affected Systems
Advertising account managers, media buyers, and agency staff using Google, Meta, TikTok, and Okta authentication. Campaign uses fake ChatGPT, Gemini, Claude, and Perplexity sites hosted on Vercel/Railway/Render infrastructure. Active since at least March 2026.
Exploitation Status
Active exploitation confirmed. Island researchers observed hundreds of victim submissions via attacker Telegram control channel. Campaign uses live operator-controlled phishing kit with real-time interaction to bypass MFA including SMS codes, authenticator apps, Okta push, and Google approval prompts.
Business Impact
Compromised advertising accounts allow attackers to spend client balances on fraudulent campaigns or resell access. Browser-in-browser (BitB) technique creates convincing fake OAuth popups that capture credentials and MFA codes in real-time. Human operators adapt attack flow based on victim responses. Traffic masking makes detection difficult as requests appear to originate from legitimate AI product APIs rather than phishing infrastructure.
Urgency
🟠 Within 24 hours
Recommended Actions
- Block known malicious domains listed in Island report at DNS/proxy layer, focusing on Vercel-hosted frontends with Railway/Render backends
- Train advertising and media buying teams to test OAuth popups by attempting to drag window outside browser or resize—BitB iframes cannot perform these actions
- Enforce hardware security keys (FIDO2/WebAuthn) for Google, Meta, TikTok, and Okta accounts used for advertising management to prevent real-time phishing
- Monitor for Socket.IO traffic patterns and Next.js stack connections to unexpected backends from advertising team workstations
- Review advertising account activity logs for unauthorized logins or spending anomalies since March 2026, especially accounts with multi-client access