Hackers Who Hit Water Utilities Were Using ChatGPT to Scout Targets

Hey everyone, Shiichan here! Today's story is about a hacking group with a track record of actually knocking out water infrastructure, and how they were using ChatGPT along the way.

OpenAI News openai.com

What was announced?

OpenAI's News page announced it had banned accounts linked to CyberAv3ngers, a group reportedly affiliated with Iran's IRGC that specializes in disruptive attacks on industrial control systems. OpenAI found the accounts were using ChatGPT for reconnaissance and development work.

Why it matters

CyberAv3ngers isn't a theoretical threat, they've caused real damage before. In November 2023 they compromised programmable logic controllers (PLCs) at the Municipal Water Authority of Aliquippa in Pennsylvania, and in December 2023 they disrupted water services in County Mayo, Ireland. Seeing exactly how a group with that track record uses an AI model for "homework" makes this a genuinely concrete case study for AI safety.

What changes

OpenAI identified and banned the accounts believed to belong to this group. That said, OpenAI's own assessment concluded the interactions provided "no novel capability, resource, or information" beyond what's already available through public sources and existing pentesting tools. In other words, this isn't a case of AI supercharging an attack, it's more like ChatGPT becoming one more research tool in the attacker's toolbox.

Dive Deep

Breaking down what OpenAI observed by phase:

• Reconnaissance: asking about default credentials for industrial devices and gathering intelligence on infrastructure in specific regions (prompts suggested interest in Jordan and Central Europe)
• Development help: getting assistance debugging bash and Python scripts
• Vulnerability research: asking about known exploits across multiple product categories
• Evasion techniques: seeking advice on code obfuscation and bypassing anomaly detection
• Post-compromise activity: asking about credential harvesting techniques after a breach

What they were actually asking for, default or weak passwords, well-documented vulnerabilities, publicly available pentesting tools, was mostly information that's already out there.

Wrap-up

• OpenAI banned accounts tied to CyberAv3ngers, a hacking group affiliated with Iran's IRGC
• The group has a real track record: compromising PLCs at a Pennsylvania water utility and disrupting water service in Ireland, both in 2023
• ChatGPT was used across the attack lifecycle: reconnaissance, scripting, vulnerability research, evasion, and post-compromise credential harvesting
• OpenAI's assessment found the information obtained stayed within the bounds of public resources and existing tools, with no novel capability gained
• Worth reading for anyone in security who wants to know how attackers targeting critical infrastructure actually use AI tools.