AI-Assisted Phishing Toolkit Uses Fake Document Shares and Hidden DNS Traffic to Target Businesses Worldwide
SOCRadar's Threat Research Unit has identified a multi-stage campaign run by a Russian-speaking threat actor. It chains three tools: XaocMail, a self-hosted phishing platform; Phantom53, a stealthy command-and-control framework; and Havoc, a post-exploitation framework. XaocMail sends personalised emails with document-sharing lures that lead to ClickFix pages, which then deliver Phantom53 installer files through Cloudflare's serverless infrastructure.
Phantom53 hides its communications by using DNS-over-HTTPS through public resolvers, including Google, Cloudflare and ControlD. It beacons out to send host information and run PowerShell commands. For targets the attacker finds interesting, an on-demand component loads Havoc directly into memory. If the active control domain is blocked or taken down, a domain generation routine that changes weekly provides a fallback. Because the infrastructure relies on legitimate services, it can be harder to spot and block.
The researchers found evidence in the code of AI-assisted development and automated testing using Claude Code task structures and prompt wrappers. Preparation began in October 2025, with phishing tests and framework development in early 2026 before the launch in August 2026. Targets span technology, finance, education, government, logistics and manufacturing, with a slight tilt toward technology.