Iranian APT Cavern Manticore Is Running a Modular Hacking Framework Built With Suspicious AI Assistance
An Iran-linked threat group has been quietly running a sophisticated modular attack framework against Israeli government bodies and IT providers, and researchers at Check Point think AI may have had a hand in writing the code.
The group, dubbed Cavern Manticore, is assessed as likely connected to Iran's Ministry of Intelligence and Security (MOIS), with possible links to Lyceum, an OilRig subgroup also known as Hexane and SiameseKitten. The targeting profile fits: Israeli public sector and tech supply chain.
The centrepiece of their operation is a .NET-based command-and-control framework that takes an unusual approach to avoiding analysis. There's no packer, no string encryption, no control-flow obfuscation. Instead, the framework's components are compiled in three different formats, each requiring a different toolchain to reverse. Analysts can't just pick a single workflow and run with it. They have to context-switch mid-investigation. It's an anti-analysis strategy by friction rather than technical complexity, and it's reasonably clever.
The framework separates core C2 communication from post-compromise capabilities. Agents connect back to operator infrastructure and pull down modules as needed, meaning each victim gets a tailored deployment rather than a one-size-fits-all implant. Available modules cover the usual post-exploitation wishlist: file operations, database enumeration, LDAP and SMB brute-forcing, network recon, and SOCKS5/WebSocket tunnelling.
Memory hygiene is also built in. Each module runs in its own isolated AppDomain and is removed from memory once unloaded, which strips out any recoverable assembly artifacts. The agent also wipes its working directory after itself, leaving behind only the communication module, config, and logs.
Initial access comes via SysAid, specifically by abusing its software update mechanism to sideload a malicious WinDirStat DLL. From there the Cavern agent executes and establishes C2 contact.
The AI angle is worth noting without overstating it. Check Point suspects the framework was partly built using an AI coding tool, but code comments, typos, inconsistent naming conventions, and structural quirks across modules all point to substantial human involvement. This wasn't just prompt-and-deploy.
On the operational side, the group has been using remote monitoring and management tools for lateral movement, browser-based remote desktop access to get eyes inside victim environments, and, in a quietly inventive move, remote printing features for data exfiltration.
Perhaps most telling is their grasp of Israel's IT supply chain. Check Point observed the group compromising an initial IT provider, pivoting to a second provider, and only then reaching the actual intended target. That kind of patience and supplier-chain awareness is not accidental. Someone did their homework.