Siemens PLC漏洞評測：AI寫攻擊腳本，NSA示警關鍵基礎設施 | Siemens S7 PLC Review: AI Exploits Hit US Infrastructure
By Kit 小克 | AI Tool Observer | 2026-09-07
🇹🇼 Siemens PLC漏洞評測：AI寫攻擊腳本，NSA示警關鍵基礎設施
AI生成漏洞攻擊已經不是理論，而是進行式。2026年8月19日，NSA、CISA、FBI、能源部（DOE）與環保署（EPA）五個美國政府機關聯合發布安全公告 AA26-231A，證實駭客正使用 AI 生成的 Python 攻擊腳本，鎖定西門子（Siemens）S7 系列可程式邏輯控制器（PLC），對美國關鍵基礎設施進行偵查與武器化開發。
什麼是Siemens S7 PLC攻擊？AI在其中扮演什麼角色
這波攻擊鎖定從 S7-200 到最新 S7-1500 F 系列安全控制器的整條產品線。攻擊者利用開源函式庫 python-snap7（封裝 snap7.dll）與 S7comm 通訊協定溝通，寫出偽裝成合法 OT（Operational Technology）監控軟體的自訂工具，藉此讀寫 PLC 的記憶體、組態資料與 ladder logic 程式。
重點是：以前寫這類工具需要相當深厚的工控協定知識，門檻不低；現在攻擊者用 AI 生成攻擊腳本，大幅壓縮了開發時間與技術門檻，等於把 ICS（工業控制系統）攻擊的「入場券」發給了更多技術能力較弱的行為者。公告用詞很直白：這「不是理論風險」。
受害產業與攻擊手法
- 目標產業：關鍵製造、能源、水與污水處理系統、化學業、食品與農業、商業設施
- 偵察方式：透過 Censys、ZoomEye 等網路掃描服務，找出暴露在公網上、韌體過舊或驗證薄弱的 PLC
- 攻擊工具：AI 生成的 Python 腳本，偽裝成 OT 監控軟體
該怎麼防？公告給的具體建議
五機關公告給出的緩解措施非常實際，不是空話：
- 盤點所有 S7 裝置，確認暴露面
- 更新 PLC 韌體到最新版本
- 在網路邊界封鎖 TCP 102（S7comm 使用的埠）
- PLC 絕對不要直接掛在公網上
- 啟用 PLC 密碼與保護等級（Protection Level）設定
- 在工程工作站以外的地方搜尋是否存在
snap7.dll、python-snap7，那不該出現在不該出現的地方 - 監控非變更窗口期間出現的 S7comm PUT/GET 操作
小克怎麼看
這次事件的意義不在於西門子 PLC 本身有多脆弱（老實說 S7comm 協定本來就沒有原生加密，這是業界公開的秘密），而在於「AI 生成漏洞」把原本需要工控背景才寫得出來的攻擊工具，變成任何懂一點 Python 的人都能組裝。對做 OT/ICS 資安的團隊來說，這是實際要處理的威脅，不是研討會投影片上的假設情境；對一般 IT 團隊來說，這也是個提醒：AI 降低攻擊門檻是雙面刃，防守方也得跟著把基本功（資產盤點、韌體更新、網路分段）補齊，因為對手已經不需要花好幾個月手刻漏洞了。
好不好用，試了才知道。
🇺🇸 Siemens S7 PLC Review: AI Exploits Hit US Infrastructure
AI-generated exploits against industrial control systems are no longer theoretical. On August 19, 2026, five US government agencies (NSA, CISA, FBI, the Department of Energy, and the EPA) jointly issued security advisory AA26-231A, confirming that threat actors are using AI-written Python scripts to conduct reconnaissance and capability development against Siemens S7 Series programmable logic controllers (PLCs) in US critical infrastructure.
What Is the Siemens S7 PLC Attack, and Where Does AI Fit In?
The advisory covers the entire S7 product line, from the S7-200 through the latest S7-1500 F-series safety controllers. Attackers are wrapping the open-source python-snap7 library (built on snap7.dll) to talk to PLCs over the S7comm protocol, producing custom tools disguised as legitimate OT monitoring software that can read and write PLC memory, configuration data, and ladder logic programs.
The key shift: writing this kind of tooling used to require deep OT protocol expertise, a real barrier to entry. Now, attackers are using AI-generated exploit scripts to compress development time and technical skill requirements dramatically, effectively lowering the entry bar for ICS attacks to a much wider pool of less-skilled actors. The advisory is blunt about it: this is not a theoretical risk.
Targeted Sectors and Attack Methods
- Targeted sectors: Critical Manufacturing, Energy, Water and Wastewater Systems, Chemical, Food and Agriculture, Commercial Facilities
- Reconnaissance method: Scanning services like Censys and ZoomEye to find internet-exposed PLCs running outdated firmware or weak authentication
- Attack tooling: AI-generated Python scripts disguised as OT monitoring software
Mitigations: What the Advisory Actually Recommends
The five-agency advisory gives concrete, actionable mitigations, not vague guidance:
- Inventory every S7 device and map your exposure
- Patch PLC firmware to the latest version
- Block TCP 102 (the port S7comm uses) at the network perimeter
- Never expose PLCs directly to the public internet
- Enable PLC passwords and Protection Level settings
- Hunt for
snap7.dllandpython-snap7outside approved engineering workstations, they should not be there - Alert on S7comm PUT/GET operations occurring outside scheduled change windows
Kit's Take
The real story here is not that Siemens S7 is uniquely fragile, S7comm's lack of native encryption has been an open industry secret for years. The story is that AI-generated exploits just turned a tool that used to require real OT/ICS background into something anyone with basic Python skills can assemble. For OT security teams, this is a live threat to operationalize today, not a conference slide hypothetical. For general IT teams, it is a reminder that AI cutting attack costs is a two-way street, defenders need to catch up on the basics (asset inventory, firmware patching, network segmentation) too, because the other side no longer needs months to hand-craft an exploit.
好不好用，試了才知道。
Sources / 資料來源
- CISA Advisory AA26-231A: Defending Against an Active Threat to Siemens S7 Series PLCs
- The Hacker News: AI-Generated Exploit Scripts Target Siemens S7 PLCs in U.S. Critical Infrastructure
- The Register: Feds warn as attackers use AI-made code to hack critical infrastructure controllers
延伸閱讀 / Related Articles
- Siri Gemini評測：蘋果砸10億美元外包AI，賭硬體翻身 | Siri Gemini Review: Apple Pays $1B/Year, Bets on Hardware
- 人形機器人出貨量評測：中國吃下97%，全年估上看5萬台 | Humanoid Robot Shipments Review: China Owns 97% in 2026
- GLM-5.3評測：MIT免費開源，Coding跑分暴增6倍 | GLM-5.3 Review: MIT-Licensed, 6x Coding Benchmark Jump
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言