The U.S. government issued a warning about an active threat campaign targeting critical infrastructure using AI-generated exploit scripts. The activity focuses on Siemens S7 Series Programmable Logic Controllers (PLCs), conducting reconnaissance and capability development. The threat actors are using AI-generated scripts disguised as legitimate monitoring tools to evade detection. The campaign represents a novel use of AI in offensive cyber operations against industrial control systems. The targeting of PLCs in critical infrastructure raises significant concerns about potential disruption to essential services. This activity highlights the growing intersection of AI technology and nation-state or sophisticated threat actor operations.
Two related AI-driven attack campaigns are described. (1) U.S. Critical Infrastructure PLC Campaign: Threat actors use internet scanning services (Censys, ZoomEye) to identify internet-exposed Siemens S7 Series PLCs running outdated or poorly protected firmware. AI-generated Python exploitation scripts, leveraging open-source industrial automation libraries such as 'snap7.dll' and 'python-snap7', are disguised as legitimate monitoring tools and communicate via the S7comm protocol, providing read/write access to PLC memory, configuration data, and ladder logic programs. Objectives include initial access, credential access, and denial of service. The use of AI lowers technical barriers for ICS attacks by accelerating script generation and iteration. Targeted sectors include Critical Manufacturing, Energy, Water and Wastewater Systems, Chemical, Food and Agriculture, and Commercial Facilities. (2) Multi-Agent Autonomous Attack on Taiwan (July 1-4, 2026): A near-autonomous AI-powered attack framework built on 'Hermes' and 'OpenClaw' agents deployed up to eight parallel lettered sub-agents (A, B, C, D, E, F, I, Q), each targeting different attack surfaces: SSO exploitation, JWT bypass, CAPTCHA brute-force (using Tesseract OCR), reconnaissance across government portals, API scanning, admin panel bypass, CVE research, vulnerability chain testing, supply chain assessment, and deep API endpoint exploitation. The framework discovered hidden API endpoints returning valid authenticated sessions regardless of request body, harvested employee usernames, cracked 85 government accounts via password spraying, exfiltrated 2,564+ personnel records, seven SSO client secrets, six internal database credentials (MSSQL, Oracle, Sybase), and internal network IP ranges. It also expanded to government IT supply chain vendors, a nuclear safety agency, a government email system, and 7+ energy sector companies. A learning engine queried vulnerability databases, GitHub repositories, and security research to adapt techniques. In roughly four days, the framework produced 1,395 files and gained persistent backdoor access. The campaign is attributed to a likely Chinese-language operator.
1. Ensure all Siemens S7 Series PLCs and other PLC devices are running the latest firmware and software versions. 2. Isolate PLCs and OT systems from the internet wherever possible; implement network segmentation between IT and OT environments. 3. Enforce strong access controls on all PLC and ICS devices, including multi-factor authentication where supported. 4. Deploy security monitoring tools to detect anomalous or malicious activity in ICS/OT environments, including unusual S7comm protocol traffic. 5. Block or restrict access to internet scanning services targeting PLC infrastructure. 6. Monitor for use of open-source industrial automation libraries (snap7.dll, python-snap7) in unexpected contexts. 7. Review and harden API endpoints to ensure authenticated sessions are not returned regardless of request body. 8. Implement CAPTCHA and account lockout policies to mitigate password spraying and brute-force attacks. 9. Conduct supply chain security assessments for government IT vendors and connected critical infrastructure entities. 10. Refer to the joint advisory (CISA AA26-231A) published by NSA, CISA, FBI, DOE, and EPA for additional guidance.
Custom Python scripts mimicking legitimate PLC monitoring tools using snap7.dll or python-snap7, S7comm protocol traffic to/from internet-exposed PLCs from unexpected sources, Use of Censys and ZoomEye for PLC reconnaissance, AI agent names: OpenClaw, Hermes, Tesseract OCR used for CAPTCHA bypass, Password spraying activity against government portals, Hidden API endpoints returning authenticated sessions regardless of request body, Exfiltration of personnel records, SSO client secrets, and database credentials, Persistent backdoors installed on government web applications, Parallel sub-agent activity (agents labeled A, B, C, D, E, F, I, Q)
