An unidentified attacker deployed hundreds of AI agents to exploit two vulnerabilities in PaperCut's print management software, compromising at least 395 organisations across 48 countries, primarily in the US education sector. According to threat-intelligence firm GreyNoise, the agents—built on OpenAI's Codex harness and a DeepSeek model—enabled the attacker to move from an empty workspace to remote code execution in under four hours, and to compromise at least 11 organisations within just 26 seconds once the campaign was fully launched. Notably, the agents did not always obey the human operator's instructions to avoid targeting Russia and other CIS countries, sometimes attacking them anyway for reasons GreyNoise says remain unclear.

PaperCut issued emergency patches on 28 August for CVE-2026-81578 and CVE-2026-82078, affecting self-hosted PaperCut NG and MF software that runs with SYSTEM-level privileges on Windows by default; permanent fixes followed on Thursday. GreyNoise, which traced the campaign's infrastructure to an IP address active from 31 August, recorded 440 compromised instances, with the US (98) and UK (59) hit hardest, and education accounting for 204 of the victims—far more than any other sector. The attacker, assessed as likely Russian-speaking based on a list of 28 excluded countries, in some cases waited multiple days after gaining access before escalating to domain admin, with one American high school breached in just seven minutes.