Researchers at Blackpoint Cyber and GreyNoise documented a campaign in which a suspected Russian-speaking attacker deployed hundreds of AI agents — built on OpenAI Codex and a DeepSeek model — to automatically exploit two vulnerabilities in the PaperCut NG/MF printing software, compromising more than 440 instances across 395 organizations in 48 countries.
According to the report, the attacker first built a private lab with a vulnerable copy of PaperCut NG/MF and an Active Directory server to develop and test exploits against two specific flaws: CVE-2026-81578 (authentication bypass) and CVE-2026-82078 (remote code execution). With the exploits ready, the attacker used the internet-scanning service Netlas.io — via an API key identified by researchers — to build target lists at global scale. The earliest detected activity dates back to August 31, 2026.
What sets this campaign apart from a traditional automated attack is the use of infrastructure specifically built to coordinate AI agents in parallel: researchers identified the use of "Hindsight," a persistent memory service for agents, and "AionUi," a unified graphical workspace that runs multiple agents concurrently. On top of that base, the agents — powered by the OpenAI Codex harness combined with a DeepSeek model — automated everything from initial reconnaissance through post-exploitation: credential harvesting, extracting Windows registry hives, lateral movement using well-known offensive tools like Mimikatz, SharpHound, Certipy, Rubeus, and Impacket, and deployment of Java-based Meterpreter payloads.
The result: credentials stolen from 280 victims, OS-level or domain secrets obtained from 147, and domain administrator privileges achieved in 12 organizations. Blackpoint Cyber notes the behavior pattern matches "initial access" activity — meaning the attacker appears to be establishing entry points at scale without a clear final objective yet, whether that's selling access, deploying ransomware, or data theft.
For any company in Guatemala or Latin America running PaperCut NG/MF — very common in educational and corporate institutions for print management — the first step is checking patch status against CVE-2026-81578 and CVE-2026-82078 and reviewing logs for recent anomalous activity. But the broader point for any security or AI lead at a company is that the same agentic automation promising productivity gains in development and operations is also drastically cutting the cost and time an attacker needs to scale an offensive campaign. Investing in enterprise AI without investing in defensive parity — monitoring, anomalous-behavior detection, accelerated patching — leaves an asymmetry wide open that attackers are already exploiting actively.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel