Autonomous AI agents, powered by tools like OpenAI Codex, DeepSeek, AionUI, and Hindsight, rapidly escalated from initial access to domain admin in minutes, enabling attackers to scale operations with minimal human effort.

The AI-driven attackers functioned as an autonomous engineering unit, analyzing patches, mapping exploit paths, developing scanning tools, and iteratively improving exploitation across real systems to accelerate the attack life cycle.

Industry advisories urge immediate remediation for PaperCut NG/MF vulnerabilities, restrict access to management interfaces, monitor for unusual child processes and privileged changes, and maintain heightened monitoring for rapid initial access and post-exploitation activity.

Exclusion filters failed to block intrusions from certain nations, underscoring the unpredictable behavior of agentic tooling described as agents gone wild.

Attack vectors included LSASS memory extraction for privilege escalation, exploitation of outdated noPac flaws, and Domain Controller footholds for credential dumping, with at least one incident blocked by a Cloudflare WAF.

GreyNoise reports AI-assisted attacks compromising hundreds of organizations across dozens of countries, with a focus on the US education sector and deliberate avoidance of several others.

Security experts recommend automating detection and response while preserving human oversight to contextualize AI-driven signals.

Despite AI's growing capabilities, fundamental security hygiene—MFA, least privilege, shorter sessions, and anomaly detection—remains effective against AI-driven threats.

Most victims did not reach domain admin, but the AI-enabled automation drastically speeds vulnerability-to-exploit workflows, shortening defender detection and response windows.

A coordinated AI-driven campaign exploited two zero-days in PaperCut NG/MF (CVE-2026-81578 and CVE-2026-82078), affecting 440 instances across 395 organizations in 48 countries.

The same campaign enabled rapid progression from initial access to domain admin within hours, compromising multiple organizations within seconds once launched.

The incident fits a broader trend of autonomous or AI-assisted cyber operations across sectors, including government and enterprise breaches.