Attackers used an AI agent to dig deeper after breaking in through PHP2Shell, Palo Alto Networks says.

The PHP flaws known as PHP2Shell, patched on Oct 21, are now part of real break-ins. Palo Alto Networks' Unit 42 research team says attackers who got into a European government entity through the flaws handed the follow-up work to "agentic AI attack platforms," and within two hours they had reached password hashes and multi-factor authentication secrets.

The attackers also set up an NTLM-relay tunnel into the organization's Active Directory and tried to get Domain Admin rights (Unit 42, updated Oct 23). What gave the agent away was formatting: each batch of commands was wrapped in headers like ==SHELLUSER== and ==PASSWD== that produce "machine-parseable output for an AI model to read," along with short tasks run one after another and quick changes of approach after failures. The Hacker News also reports that security firm CYFIRMA has seen a Korean-speaking group, which it links to North Korea's Lazarus hackers with moderate confidence, quietly scanning a shortlist of 216 unpatched servers at governments, banks, airports and other targets (The Hacker News).

Unit 42 counts more than 22.6 million exposed PHP instances that may be vulnerable.

What this means for you: if you run a site on PHP, update to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 (or your host's or Linux distribution's patched package) and restart PHP. The PHP project says there are no complete workarounds; it also now suggests setting session.upload_progress.enabled=0 until the next regular release to close a third, unlikely path (PHP).