NewsAnalysis
LiteLLM's postmortem: a master key committed to a public repo in May and never revoked opened the door to the October release
LiteLLM published the attack chain behind its compromised 1.107.0 release on October 23, and The Register summarized it on October 26. On May 27 a configuration file containing a master key for an open source tool LiteLLM used internally to orchestrate AI agents was committed to the public LiteLLM-Labs/lite-harness repository; developers removed the commit quickly but did not revoke the key. On October 10 the attacker used the key to reach the LiteLLM Agent Platform and extract a GitHub CLI OAuth token with owner-level access to every repository in LiteLLM's GitHub organization, a post-2FA token that does not trigger GitHub's sudo mode for high-impact actions. From there they committed malicious code to branches in other internal repositories, including workflows built to steal CI tokens and secrets, took the GitHub CLI OAuth tokens of two engineers with maintainer privileges and tokens for several other internal systems, with a focus on upstream inference API tokens, all under the names of credentialed engineers. The trojanized package was on PyPI for under two hours and, by Wiz's count, was downloaded about 49,000 times. LiteLLM is revoking every token tied to every internal system and vendor platform and calls the incident a strong illustration of the risk of failing to rotate a potentially exposed key.
Why it matters
Removing a commit does not unpublish a secret. Any key that touched a public repository, a log or a chat is burned the moment it lands there, whatever happens next; rotate it that day, and treat tokens that skip re-authentication as the crown jewels they are.