Translation in progress Some entries still include text in the original language.
BackSource CSA Labs Research
CSA Labs Research
Indexed by this site 2026-10-24 21:15 6 days ago
Originally published Oct 24
Selected 2026-10-24 21:16
Category Event CSA Labs Research · Indexed by this site 2026-10-24 21:15 · Originally publishedOct 24 SelectedAttention54
The unrevoked key: from agent credentials to LiteLLM release hijacking The Unrevoked Key: Agent Credential to LiteLLM Release Hijack
AI summary
2026年10月21日,携带凭据窃取载荷的 litellm==1.107.0 被发布到 PyPI 和 LiteLLM 官方 Docker Hub 与 GHCR 镜像,这是 LiteLLM 2026年内第二次供应链投毒。CSA 依据 LiteLLM 的攻击链披露梳理:一枚 5月27日提交到公开仓库、从未撤销的 Agent Platform master key 在 10月10日被用于调取内部 API,返回一位 GitHub 组织所有者的 CLI OAuth token;攻击者借此在11天里收集多名维护者 token,关闭分支保护并把发布工作流指向仿冒仓库 Berri-AI/litellm,由正规管道构建、签名并发布了带有效溯源的恶意制品。载荷在每次 Python 启动时运行窃取器,转储代理数据库中的上游提供商密钥,并启动每小时轮询远程代码的守护进程。
Why it was selected
The article lays out how an attacker used a single unrevoked key to drive the legitimate release pipeline to complete the poisoning, with neither signing nor provenance stopping it, which can be used to check one's own agent credential governance.
Reported by another 17 sources Read original labs.cloudsecurityalliance.org
Why it was selected SelectedAttention54 2 new sources in 24 hours
Attention 0–100: content value contributes up to 50 and circulation up to 50 (accounts weighted by likes); circulation decays over time. How it was rated
The article lays out how an attacker used a single unrevoked key to drive the legitimate release pipeline to complete the poisoning, with neither signing nor provenance stopping it, which can be used to check one's own agent credential governance.