GreyNoise: AI agents hit 395 organizations through PaperCut

GreyNoise published a lab note describing an AI-orchestrated campaign against PaperCut print-management software that it says compromised at least 440 instances at 395 organizations.

A security-monitoring firm published a dated case of AI agents used in an attack, not a keynote demo. Patch PaperCut. The victim counts and timings are GreyNoise’s telemetry.

GreyNoise published “Agents Gone Wild.” It describes a campaign that began 31 Aug 2026 against PaperCut NG/MF, a print-management product. The bugs are listed as CVE-2026-81578 and CVE-2026-82078 — Common Vulnerabilities and Exposures identifiers, the public catalog numbers for those bugs.

GreyNoise: a likely Russian-speaking actor used hundreds of AI agents — OpenAI’s Codex harness plus a DeepSeek model, not OpenAI models — and public offensive tools. It says they compromised at least 440 PaperCut instances at 395 identified organizations in 48 countries.

GreyNoise timings: empty workspace to first real-victim remote code execution — running the attacker’s code on the victim’s machine — in under four hours. First domain-admin two hours later, meaning control of the Windows domain. Once the full wave launched, at least 11 organizations in 26 seconds. A U.S. high school went from initial access to domain admin in seven minutes.

GreyNoise: domain-admin on only 12 organizations; credentials from 280; OS or domain secrets from 147. Education accounted for 204 victims, which GreyNoise ties to PaperCut’s installed base in schools.

The actor listed 28 countries to avoid. GreyNoise says some of those countries were hit anyway — the “agents gone wild” line. Cloudflare’s WAF (web application firewall, a filter in front of a website) stopped at least one attempt.