AI Agent Swarm Exploits PaperCut CVEs to Breach 440 Instances in 48 Countries

The actor built a lab with vulnerable PaperCut and Active Directory, then used Netlas.io API scans to generate target lists while testing authentication bypass plus RCE. Post-exploitation delivered registry hive tools, Meterpreter Java payloads, Mimikatz, SharpHound, Certipy, Rubeus, and Impacket. Domain administrator access was achieved in 12 organizations, including a U.S. high school where full compromise occurred in seven minutes. GreyNoise telemetry shows the IP probed Palo Alto, Citrix, and SonicWall instances since July 2026; Blackpoint and Arctic Wolf independently linked the same address to the PaperCut activity. The actor explicitly excluded 28 countries including Russia and China yet still struck victims in several excluded jurisdictions, indicating control failures rather than disciplined targeting. Technical evidence is limited to IP correlation, tool hashes, and agent-generated commands; no malware samples or C2 infrastructure have been publicly shared to confirm state affiliation. The speed from empty workspace to 11 organizations in 26 seconds demonstrates agentic AI collapsing reconnaissance-to-execution timelines previously measured in days. Initial access broker activity remains the working hypothesis, but the same infrastructure could pivot to ransomware or data exfiltration. Defenders should audit PaperCut instances for the two CVEs and monitor for anomalous registry collection or SharpHound activity within the next 30 days.