AI agents exploit PaperCut vulnerabilities in 395-organization cyberattack
A single attacker used AI agents powered by commercial models to compromise 440 servers at 395 organizations worldwide by exploiting unpatched vulnerabilities in PaperCut print management software. The campaign affected mostly schools and universities, with the attacker obtaining domain administrator access in some cases within minutes. According to threat intelligence firm GreyNoise's analysis, the attacker compressed tasks normally requiring specialist teams into hours of automated exploitation.
What happened
On September 9, 2026, threat intelligence firm GreyNoise published findings that a single attacker used autonomous AI agents powered by OpenAI's Codex and a DeepSeek model to compromise 440 servers belonging to 395 organizations across 48 countries, predominantly schools and universities.
The attacker exploited two vulnerabilities in PaperCut NG and MF (CVE-2026-81578 and CVE-2026-82078): an improper access control flaw allowing unauthenticated modification of system configurations, and an unsafe dynamic class-loading flaw enabling arbitrary Java code execution. The attacker built, tested, and deployed the exploit swarm after gaining remote-code execution in less than four hours from an empty workspace.
The attacker achieved 11 compromises within 26 seconds at the campaign's peak. At one U.S. high school, domain administrator access was obtained within seven minutes of initial access. The attacker obtained domain administrator access at 12 organizations total.
Reported impact
At organizations where the attacker gained administrator access, domain credentials and Windows domain credentials were extracted. PaperCut's default configuration running with SYSTEM-level privileges and integration with Active Directory meant compromised servers provided access to every account in the organization.
Index analysis
The attacker used the Netlas search service to identify internet-facing PaperCut installations as targets. GreyNoise observed the attacker develop and test exploits in a lab containing vulnerable PaperCut software and an Active Directory server.
The attacker deployed hundreds of AI agents using OpenAI's Codex as a harness along with a DeepSeek model and publicly available offensive security tools. The compression of tasks that would have required a specialist team into a single attacker's workflow in hours demonstrates the speed and scale possible with AI-assisted exploitation.
Governance controls
PaperCut published emergency fixes in late August 2026 after receiving reports of customer compromises, with updated versions released on September 10. The vulnerabilities were listed in the CISA Known Exploited Vulnerabilities catalog.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.