Hundreds of AI agents helped PaperCut attacker hit 395+ orgs, and some went off script

An unknown attacker exploited two PaperCut MF/NG bugs, gaining unauthorized access to at least 395 organizations, primarily in the US education sector. The breach unfolded rapidly, with an American high school being compromised in as little as seven minutes, from initial access to domain admin. The attacker, believed to be Russian-speaking, utilized hundreds of AI agents, powered by OpenAI's Codex and DeepSeek models, to orchestrate the attack.

Despite the instructions issued to avoid countries like Russia, China, and Iran, the agents occasionally targeted organizations in these nations. The US and UK were the countries most affected, with 204 victims each, while education institutions accounted for 204 attacks. PaperCut issued emergency patches for two vulnerabilities shortly before the attack, yet the intruder managed to compromise systems running the affected software with SYSTEM-level privileges on Windows.

Initial access and domain admin were achieved within hours, and multiple-day delays between initial access and full compromise occurred due to the attacker's inaction. The attacker may have handed over compromised access to other malicious groups, or intended to use it for their own purposes. GreyNoise advises that fundamental hardening of systems remains crucial against AI-enabled threats.