Agentic Flooding評測：AI代理灌爆各國公部門 | Agentic Flooding Review: AI Agents Swamp Gov Services

9月 12, 2026

By Kit 小克 | AI Tool Observer | 2026-09-12

🇹🇼 Agentic Flooding評測：AI代理灌爆各國公部門

什麼是 Agentic Flooding（AI代理灌爆）？

Agentic Flooding 這個新名詞這週開始在 AI 圈子裡發酵——研究員 Chris Schmitz 在一份即將於 AI Ethics and Society 研討會發表的論文中，記錄了 11 個國家、84 起因為 AI 代理而爆量的公部門申訴案例。簡單說，就是當 AI代理 把「寫一份陳情信」「填一張申訴表」的成本壓到接近零，原本靠「麻煩」擋住的申請量一下子全部湧進來，公部門的人力和系統根本吃不消。

數字有多誇張：從2,600到破7,000

這不是紙上談兵，是已經在發生的事。英國住房監察使（Housing Ombudsman）的申訴案件，從 2022 年的 2,600 件暴增到 2025 年超過 7,000 件——時間點剛好對上 ChatGPT 普及；美國消費者金融保護局（CFPB）同期申訴量更是翻了 5 倍。巴西司法系統的陳情案、德國聯邦議院的請願案也都出現類似的暴增曲線。這些案例橫跨政府申訴、司法陳情、國會請願等不同體系，說明這不是單一國家或單一部門的個案，而是 agentic flooding 這個現象的全球擴散。

為什麼稱它是「程序性DDoS」

傳統官僚體系其實靠「摩擦力」在自我保護——填表要花時間、要懂法律用語、要付出心力，這道門檻天然把申請量壓在人力可處理的範圍內。AI 代理一來，這道摩擦力直接被抹平：任何人只要打幾句話，AI 就能幫你生出一份格式正確、用詞專業的申訴信。研究者把這種現象比喻成「程序性 DDoS 攻擊」——不是駭客惡意攻擊伺服器，而是大量本來會放棄申請的「合法使用者」，因為 AI 代理降低了門檻而一起湧入，系統一樣會被灌爆。值得注意的是，論文也承認：這些暴增裡有一部分確實是惡意濫用，但更多其實是原本會放棄申請的民眾，靠 AI 代理才把該爭取的權益爭取回來。

對開發者與一般用戶的啟示

做 AI 代理產品的人：任何會生成「送出去給人審核」內容的代理工具，都該想清楚下游承接單位的處理量能，不是做出來能跑就上線。
一般用戶：用 AI 代理幫你寫申訴信本身沒問題，但別指望「量多」就能加速結果——公部門的隊伍只會更長，不會更快。
政策制定者：這波趨勢代表傳統「靠摩擦力限流」的行政設計已經失效，需要重新思考審核流程的量能規劃。

好不好用，試了才知道。

🇺🇸 Agentic Flooding Review: AI Agents Swamp Gov Services

What Is Agentic Flooding?

"Agentic flooding" is a new term working its way through the AI world this week. Researcher Chris Schmitz documented 84 cases across 11 countries in a paper heading to the AI Ethics and Society conference, all describing the same pattern: AI agents drive a surge in filings to public services. Once an agent can draft a complaint letter or benefits appeal for free, the natural friction that used to keep filing volume manageable disappears — and queues built for human-scale traffic get hit with agent-scale traffic instead.

The Numbers: From 2,600 to Over 7,000

This isn't theoretical. The UK's Housing Ombudsman saw complaints jump from 2,600 in 2022 to more than 7,000 in 2025 — right in step with ChatGPT's mainstream adoption. The US Consumer Financial Protection Bureau recorded a 5x increase in complaints over the same window. Brazil's judiciary and Germany's federal parliament (Bundestag) both logged similar spikes in petitions. The pattern shows up across completely different systems — housing, finance, courts, legislatures — which is what makes agentic flooding look like a structural shift rather than a one-off glitch.

Why Researchers Call It a "Procedural DDoS"

Bureaucracy has always relied on friction as a load-bearing feature: the time, legal literacy, and effort a form demands naturally caps how many people file. AI agents erase that friction — anyone can now produce a properly formatted, professionally worded complaint with a few sentences of prompting. Researchers describe the result as a "procedural DDoS": not malicious hacking, but a flood of legitimate requests arriving all at once because the cost of filing dropped to near zero. The paper is careful to note some of this volume is adversarial abuse, but a meaningful share is people who would have simply given up on a legitimate claim without an AI agent's help.

What This Means If You Build or Use Agents

Builders: if your agent generates anything meant for human review — complaints, appeals, applications — check whether the receiving institution can actually absorb the volume before you ship.
Users: filing faster with an agent doesn't mean a faster outcome — the queue on the other side is only getting longer.
Policymakers: intake systems designed around friction-as-a-filter no longer work once agentic flooding is in play; capacity planning needs a rethink.

好不好用，試了才知道。

Sources / 資料來源

延伸閱讀 / Related Articles

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

搜尋此網誌

AI小貼士