跳到正文
DEV Community· threataft·· 8 天前AI 評分70

FBI 的 ARTEX 警示:高噪音不再代表攻擊者技術粗疏

The FBI’s ARTEX Flash: When Noise Stopped Meaning Unsophisticated

AI 導讀

FBI 就 ARTEX 類 AI 攻擊框架發出警示,指出語言模型改變了已知攻擊手法可由誰執行及其速度,因此高噪音流量不再必然代表攻擊者技術粗疏。GitGuardian 測試一個未包含機密資料掃描程式碼的 ARTEX 版本,以只要求評估程式碼庫安全狀況、沒有提及憑證的中性提示詞進行測試,仍找到 9 個植入憑證中的 8 個,首個搜尋命令在任務開始後 6–13 秒執行。

正文

threataft

The FBI issued a flash on ARTEX-class AI attack frameworks. It doesn’t describe a new vulnerability or a specific victim set. It describes a shift in who can run known techniques, and how fast those techniques run when a language model is executing them.

The number: GitGuardian tested an ARTEX build that shipped no secret-scanning code. Prompt was neutral — assess the repo’s security posture, nothing about secrets. The agent found 8 of 9 planted credentials. First hunting command ran 6–13 seconds into the task.

The broken assumption: for two decades, SOC triage has relied on a correlation: noisy, broad, high-volume traffic signals an unsophisticated actor. Sophisticated actors move quietly. AI tooling breaks the correlation without breaking the noise — the agent fires API calls at inference-loop speed and doesn’t know it’s supposed to be quiet.

What changed: AntiFox (ARTEX 0.1.1 fork, v2.0) generated hundreds of AWS API calls in ~30 minutes in confirmed intrusions. Most returned AccessDenied. The successful GetSecretValue against a production secret was embedded in that burst. A triage analyst who closed the ticket on the denial count missed the intrusion.

What didn’t change: least privilege, IMDSv2, CloudTrail data events, OIDC instead of long-lived keys, MFA on privileged accounts — every one of those controls still stops it. What changes is which signal you treat as the incident.

A wall of denials with one success is the search. The success is the intrusion.

Full article with hunt guidance and mitigation checklist at the link above.

來源:DEV Community · dev.to