Overview
PaperCut has released full maintenance releases for its NG/MF print management software to replace a series of emergency patches addressing two actively exploited vulnerabilities: CVE-2026-81578 (authentication bypass) and CVE-2026-82078 (arbitrary code execution). Versions 26.0.5, 25.0.13, and 24.1.10 are now available and supersede all prior emergency patch releases, incorporating additional security hardening and standard QA testing.
What elevates this incident beyond a routine patch cycle is the confirmed use of AI agent infrastructure to automate exploitation at scale — a signal of a maturing AI-enabled threat landscape that warrants specific attention from the security community.
Technical Analysis
The two CVEs enable a chained attack path: CVE-2026-81578 allows an unauthenticated attacker to bypass access controls on exposed PaperCut instances, while CVE-2026-82078 enables subsequent arbitrary code execution on the compromised host.
According to GreyNoise and Blackpoint Cyber, a suspected Russian-speaking threat actor weaponised both flaws using hundreds of AI agents built on OpenAI’s Codex harness and a DeepSeek model. These agents automated target identification, exploitation sequencing, and likely evasion logic — enabling a single actor or small group to breach 395 organisations in 48 countries from a single IP address (45.142.193[.]132).
Notably, the campaign deliberately avoided targeting entities in Russia, China, Hong Kong, Thailand, Iran, and 23 other countries — a common indicator of state-adjacent threat actor tradecraft.
Framework Mapping
MITRE ATLAS:
- AML.T0103 – Deploy AI Agent: The threat actor deployed autonomous AI agents to conduct exploitation operations at scale, replacing manual attacker workflows.
- AML.T0047 – AI-Enabled Product or Service: OpenAI Codex and DeepSeek were used as enabling infrastructure for the attack campaign.
- AML.T0043 – Craft Adversarial Data: Agents likely crafted and adapted exploit payloads dynamically to target diverse PaperCut configurations.
OWASP LLM Top 10:
- LLM08 – Excessive Agency: The deployment of autonomous LLM-powered agents conducting offensive cyber operations without human-in-the-loop oversight exemplifies the systemic risk of unconstrained AI agency in adversarial hands.
Impact Assessment
The campaign’s reach — 395 organisations across 48 countries — demonstrates the force-multiplier effect of AI-assisted exploitation. The U.S. education sector bore the heaviest concentration of victims, likely due to the sector’s historically slower patch cadence and broad PaperCut adoption.
GreyNoise assesses it remains unclear whether the actor’s objective is access brokering (selling compromised footholds to ransomware affiliates) or direct follow-on action such as data theft. Either scenario carries significant downstream risk for affected institutions.
Mitigation & Recommendations
- Patch immediately: Upgrade to PaperCut NG/MF versions 26.0.5, 25.0.13, or 24.1.10. Emergency patch builds should be migrated to full maintenance releases.
- Block known malicious infrastructure: Add
45.142.193[.]132to network-level blocklists and monitor for related traffic patterns. - Review authentication logs: Inspect PaperCut server logs for anomalous login attempts, unexpected admin account creation, or unusual API activity.
- Restrict external exposure: Where possible, limit PaperCut admin interfaces to internal networks or VPN-only access.
- Monitor for follow-on activity: Given the ambiguity around actor intent, deploy endpoint detection rules for signs of lateral movement, credential dumping, or data staging post-compromise.