CISA AA26-231A: AI wrote the exploit, you left the port open

Five federal agencies signed a joint advisory this week saying threat actors are using AI to write exploitation scripts against Siemens S7 programmable logic controllers in US critical infrastructure. The interesting part of CISA AA26-231A is not the AI. It is that the AI only mattered because the devices were reachable from the internet in the first place.

What actually happened

NSA, CISA, FBI, the Department of Energy, and the EPA published advisory AA26-231A on August 19, 2026, covering the Siemens S7 series across generations — S7-200, S7-300, S7-400, S7-1200, and S7-1500.

The attack chain is unglamorous, which is the point. Actors use internet scanning services — Censys and ZoomEye are named — to find exposed PLCs running outdated software. They then pair publicly available industrial automation libraries, specifically snap7.dll and python-snap7, with AI-assisted scripting to produce custom tools that impersonate legitimate OT monitoring software. Those tools read and write PLC memory, configuration data, and ladder logic over the S7comm protocol, per reporting from BleepingComputer.

The activity observed so far is reconnaissance and capability development, not disruption. Targeted sectors include critical manufacturing, energy, water and wastewater, chemical, food and agriculture, and commercial facilities. The advisory's priority directive is blunt: get internet-exposed S7 PLCs off the internet, patch what remains, tighten access controls, and monitor for anomalous ICS traffic.

Why an AI-assisted exploit advisory matters for your business

You probably do not run a PLC. You do run the same pattern.

AI collapsed the gap between a known weakness and a working tool. Writing a credible S7comm client used to require someone who understood industrial protocols. Now it requires a library, public documentation, and a model. The advisory's own framing — known vulnerabilities plus accessible libraries plus AI-assisted development equals a high-probability attack — applies verbatim to your Redis instance with no password, your staging environment on a public IP, and the admin panel you left on /wp-admin behind one password. The skill floor for turning "known issue" into "running exploit" dropped to roughly zero.

Scanning is not the attacker's cost anymore. Censys and ZoomEye are commercial services. Anything you expose is enumerated continuously by default, indexed, and searchable by version string. The question is never whether you will be found. It is what is standing there when you are.

The fix is inventory, and nobody wants to own it. Every mitigation in that advisory is a thing you were supposed to already know: what you run, what version, what is listening, who can reach it. That is a boring quarterly job with no demo. It is also the only control that scales when exploit development stops being a bottleneck.

Sources: CISA advisory AA26-231A, BleepingComputer.