It’s not hypothetical anymore. Huntress caught a threat actor using AI-generated malware in an actual network intrusion.
The tool was a PowerShell script labeled “100% Working AD Information Gathering Script – FULLY FIXED.” The hallmarks of LLM assistance were obvious: a placeholder server name, over-engineering with multiple fallback methods, and console output with excessive colors. Classic vibe-coded malware.
Here’s what it did. The attacker used stolen credentials to log in via RDP, staged tools in a common Windows folder, then ran the AI-generated script to map out the Active Directory environment. It harvested users, computers, groups, and trusts, compiled everything into spreadsheets, and even generated an HTML report. Data exfiltration was handled through legitimate cloud tools.
The intrusion itself isn’t novel. The attack pattern is familiar. What’s different is the tooling. Traditional signature-based antivirus can’t catch this because the script is one-off — it’s not a known malware sample. It was generated fresh for this operation.
This is the problem defenders are going to face more often. AI lets even low-skill attackers produce unique payloads that bypass signature detection. The answer isn’t better signatures — it’s behavioral analytics. Watch what the script does, not what it’s named.
