当AI降低工业攻击门槛：生成式漏洞利用时代的OT防御

When AI Lowers the Bar for Industrial Attacks: Defending OT in the Age of Generated Exploits For years, the security of industrial control systems rested on an implicit barrier: attacking a programmable logic controller (PLC) required deep, specialized knowledge of proprietary protocols and hardware. In August 2026, five U.S. federal agencies issued a joint advisory confirming that this barrier is eroding. Threat actors are using AI-generated code to build exploitation scripts against Siemens S7 series PLCs deployed at water, energy, chemical, and manufacturing sites. The advisory is a milestone because it is the first government alert to state plainly that AI is being used to target operational technology (OT). Why the reader should care OT systems control physical processes. A compromised PLC can mean a halted production line, a tampered process parameter, or in the worst case a safety incident. If AI collapses the expertise required to attack these systems, then the population of capable attackers grows, and the old assumption that OT is protected by obscurity no longer holds. Technical context The advisory (AA26-231A), co-signed by the NSA, CISA, the FBI, the Department of Energy, and the EPA, describes a specific technique. Attackers use internet scanning services such as Censys and ZoomEye to find Siemens S7 PLCs with port 102 exposed, then feed that targeting data into AI tools that generate working Python exploitation scripts using the open-source snap7 and python-snap7 libraries. Those scripts communicate over the S7comm protocol, which historically lacked authentication on many devices. The advisory covers the S7-200, S7-300, S7-400, S7-1200, and S7-1500 families. Critically, it does not describe a single new vulnerability. Siemens confirmed that the advisory points to configuration weaknesses and exposure, not a new product flaw. That distinction shapes the entire defensive response. Explanation and walkthrough The attack chain has four stages. First, discovery: scanning for exposed controllers on port 102. Second, generation: using AI to produce scripts that speak S7comm without the attacker needing to understand the protocol internals. Third, disguise: presenting the scripts as legitimate OT monitoring software so they blend into normal engineering traffic. Fourth, access: reading and potentially writing PLC memory, configuration data, and ladder logic. The reason this is hard to detect is that the traffic looks like ordinary engineering activity. There is no malware signature, because the tool is a legitimate library being used for an illegitimate purpose. The advisory names one concrete signal: unauthorized use of snap7.dll outside approved systems. A facility whose SCADA platform does not use snap7 that observes snap7 traffic on port 102 should treat it as an indicator of compromise. Defensive implications The response is about reducing exposure and controlling access, not waiting for a patch that may never come. Remove PLCs from direct internet exposure. Port 102 should never be reachable from the public internet; remote maintenance should run over an authenticated, monitored channel. Segment OT from IT. Follow IEC 62443 principles to isolate the production network from corporate and cloud environments. Inventory and classify controllers. You cannot protect devices you have not enumerated, including legacy models. Enforce strong authentication and change default credentials. Many of these devices ship with weak or absent authentication. Monitor for anomalous S7comm activity. Baseline normal engineering traffic and alert on deviations, especially from unexpected hosts. Plan for retirement of unpatchable hardware. S7-300 and S7-400 devices cannot be fixed with a patch; isolation or replacement is the only durable answer. The convergence of AI and OT is not a future scenario; the advisory treats it as present reality. The organizations that fare best will be those that treat their industrial networks as exposed by default and design their defenses accordingly. References Joint advisory AA26-231A (NSA, CISA, FBI, DOE, EPA), 2026-08-19, on AI-assisted attacks against Siemens S7 PLCs. TechTimes report on federal confirmation of AI-written exploits for Siemens PLCs, 2026-08-20. Siemens ProductCERT advisory SSB-104599. MITRE ATT&CK for ICS, Remote System Discovery (T0846), PLC-Blaster port 102 scanning.