PHP2Shell Flaws Exploited Worldwide, With Signs Attackers Are Using AI to Move Faster

Attackers are abusing two PHP2Shell vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, and appear to be using agentic AI platforms once they get inside. Palo Alto Networks Unit 42 described an attack on a European government entity. Within two hours of the first compromise, the attackers had reached password hashes and multi-factor authentication secrets. They also set up an NTLM-relay tunnel into Active Directory and tried to escalate to Domain Admin.

Unit 42 said the commands and behaviour pointed to an agentic attacker. Command output was wrapped in labelled section headers, which makes it easy for an AI model to read. The attacker also queried the domain controller to look for the administrator account and searched the host for internal systems, privileged groups and active users.

Separately, CYFIRMA reported a suspected North Korea-linked campaign using PHP2Shell to probe PHP systems from 21 compromised nodes. The scanning is slow and paced, and nodes are dropped at the first sign of blocking, so threshold-based alerts may miss it. CYFIRMA identified 216 unpatched or partially mitigated hosts as targets, including some in Australia. They span government, finance, logistics, aviation, manufacturing and online transaction platforms. CYFIRMA advises organisations in these sectors and regions to assume they are in scope.