AI駭客攻擊評測:23小時攻破100家公司,一次50美元 | AI Hacker Review: 100+ Sites Breached in 23 Hours
By Kit 小克 | AI Tool Observer | 2026-10-19
🇹🇼 AI駭客攻擊評測:23小時攻破100家公司,一次50美元
一名AI駭客只花了大約5000美元的token費用,就靠自動化AI agent攻破了美國100多家公司的網站,包括農具大廠迪爾(Deere)、家電品牌SharkNinja與引擎製造商康明斯(Cummins)。資安研究員Snorre Fagerland證實,這是第一次在真實世界確認:商用AI代理能夠從頭到尾完全自主完成複雜入侵,不需要人力介入每一步。
這次AI駭客攻擊怎麼運作?
駭客幾乎把整個入侵流程都交給AI代理執行——自己只負責設定攻擊目標和審查結果,AI agent則自動掃描漏洞、產生攻擊程式、入侵系統、再回報戰果。平均每次成功入侵耗時約23小時,成本約50美元的AI token費用。
用了哪些工具?
報告指出攻擊混用了DeepSeek與Anthropic的Claude Code,搭配一套原本給資安研究員用的開源問題解決平台Cairn。Fagerland正是在掃描網路上意外暴露的Cairn部署時,無意間發現這起攻擊活動的。
攻擊規模有多大?
- 攻擊活動從2026年9月25日持續至今
- 嘗試入侵超過1200個網站
- 成功攻破100多家公司的網站
- Deere表示已採取補救措施,未發現敏感資料外洩
- SharkNinja稱受影響網站僅含產品支援資訊,威脅已被控制
- Cummins則表示正在調查第三方代管網站可能遭攻擊一事
一般公司該怎麼防範AI駭客?
這起事件打破了「AI只能幫忙寫程式,發動攻擊還是要靠真人」的迷思。過去組一支紅隊要花大錢養人力,現在只要一個人、幾千美元token預算,就能讓AI agent自主掃描、入侵上千個目標。對企業來說,代管給第三方供應商的子網域、很少巡檢的舊系統,反而是最容易被AI agent掃到的破口,比起等漏洞公告,定期檢查異常流量模式更實際。
好不好用,試了才知道——只是這次「好用」的是壞人手上的攻擊工具,值得資安團隊提高警覺。
🇺🇸 AI Hacker Review: 100+ Sites Breached in 23 Hours
A hacker spent roughly $5,000 in AI token costs to autonomously breach the websites of more than 100 companies — including farm equipment maker Deere, appliance brand SharkNinja, and engine manufacturer Cummins — using off-the-shelf AI hacking agents. Security researcher Snorre Fagerland says it's the first real-world confirmation that commercially available AI agents can execute a complex intrusion from start to finish with no human intervention.
How Did This AI Hacking Campaign Work?
The hacker delegated nearly the entire attack chain to software — setting a target and reviewing results while the agent scanned for vulnerabilities, wrote exploit code, broke in, and reported back. A typical successful intrusion took about 23 hours and cost roughly $50 in AI tokens.
Which AI Tools Were Used?
The campaign reportedly mixed DeepSeek and Anthropic's Claude Code, running on top of Cairn, an open-source AI problem-solving platform originally built for security researchers. Fagerland stumbled onto the campaign while scanning the internet for accidentally exposed Cairn deployments.
How Big Was the Attack?
- Campaign has run since September 25, 2026
- Attempted intrusions against more than 1,200 websites
- Successfully breached 100+ companies
- Deere says it took remediation steps and found no evidence sensitive data was accessed
- SharkNinja says the affected site only held product support info and the threat is contained
- Cummins says it's investigating a possible attack on a third-party-hosted site
What Should Companies Do About AI Hacking Agents?
This breaks the assumption that AI is only useful for writing code, not launching attacks. Building a red team used to require real headcount; now one person with a few thousand dollars of token budget can point an agent at thousands of targets. For most companies, the weak spot is exactly where attention is thinnest — vendor-hosted subdomains and rarely-audited legacy sites — so watching for abnormal traffic patterns matters more than waiting for the next CVE advisory.
好不好用,試了才知道 — this time, the tool that "worked well" was in the wrong hands, which is exactly why security teams should be paying attention.
Sources / 資料來源
- Reuters: How One Hacker Used AI Agents and a Few Thousand Dollars to Break Into Scores of Sites
- Cryptonomist: AI Autonomous Hacking Campaign Breaches 100+ Company Websites
- Insurance Journal: How One Hacker Used AI Agents, a Few Thousand Dollars to Break Into Websites
常見問題 FAQ
這次AI駭客攻擊用的是ChatGPT嗎?
不是。報告指出攻擊混用了DeepSeek與Anthropic的Claude Code,外加開源平台Cairn,目前沒有證據顯示用到OpenAI的產品。
一般公司要怎麼知道自己是不是受害者?
可以檢查網站存取紀錄是否有異常高頻率、自動化的掃描或暴力嘗試流量,尤其是第三方代管的子網域與很少巡檢的舊系統。
AI公司對這次事件有回應嗎?
目前Anthropic與DeepSeek並未針對此事件單獨發表聲明,但這類濫用案例再次把agentic AI的安全防護推上討論焦點。
這跟Anthropic之前抓到的國家級AI攻擊事件一樣嗎?
不一樣。先前披露的事件被認為是國家級情報行動,這次則是單一駭客只花約5000美元token費用就辦到,門檔更低,代表更多人有能力複製類似攻擊。
延伸閱讀 / Related Articles
- GitLab AI Gateway漏洞評測:CVSS 9.9,樣板跳脫變RCE | GitLab AI Gateway RCE Review: CVSS 9.9 Sandbox Escape
- DeepSeek搜尋API評測:千次0.75美元,比Tavily便宜10倍 | DeepSeek Search API Review: $0.75/1K vs Tavily's $8
- AI代理人評測:六大銀行封殺,密碼授權誰負責 | AI Agent Review: Banks Block Rogue Agents Over Liability
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