Future TechnologyOne threat actor used hundreds of AI agents to compromise 440 print servers
- AI agents reached remote code execution against live targets in under four hours, then domain admin two hours later
- Educational institutions made up 204 of 440 compromised servers across 48 countries
- The attacker combined OpenAI Codex and DeepSeek models with commodity offensive tools to automate the entire kill chain
From empty workspace to domain admin in six hours
Four hundred and forty PaperCut servers across 48 countries fell in a single campaign. The attacker was one Russian-speaking threat actor armed with hundreds of autonomous AI agents built on OpenAI's Codex and a DeepSeek model.
On August 31, the operation targeted two critical vulnerabilities in PaperCut NG/MF print management software: CVE-2026-81578, an authentication bypass, and CVE-2026-82078, a remote code execution flaw via unsafe reflection. According to GreyNoise, which had been tracking the attacker's IP since early July, the campaign went from an empty workspace to remote code execution on a live victim in under four hours. Domain administrator privileges followed roughly two hours later.
The speed at scale is the part worth sitting with. In one burst, 11 organisations were compromised in 26 seconds.
Schools took the worst of it
Educational institutions accounted for 204 of the 440 compromised systems. Schools and universities tend to run older software on tighter budgets, which makes them exactly the kind of target an automated campaign can sweep through without resistance.
The United States had the largest share of victims at 98, followed by the United Kingdom, France, and Spain. PaperCut is widely deployed in education, healthcare, and government, sectors where print management is still part of daily operations.
AI agents as an offensive weapon
Automated exploitation is not new. But the PaperCut campaign went further. The AI agents wrote exploit code, adapted payloads, and escalated privileges across networks without human intervention at runtime.
The combination of foundation models with commodity offensive tools turns what used to be a slow, manual post-exploitation process into something that scales like cloud infrastructure.
What to do now
PaperCut has released patches for both CVEs. If you run PaperCut NG/MF, the fix is updating to the latest version and auditing your access logs from late August onward. The broader problem is harder to patch: as AI agents get better at chaining exploits autonomously, the window between vulnerability disclosure and mass exploitation is collapsing from days to hours.
