← Back to Tech · Phones & Mobile
a developer’s dark desk at night, a laptop showing an abstract stack of glowing package boxes with one box cracked open and leaking red light, a tiny silhouetted figure slipping a key out of it, cool blue monitor glow, no logos, no text

LiteLLM 1.107.0 was hijacked to steal developers’ credentials

A second security story for builders today. LiteLLM, a popular open-source tool developers use to connect apps to AI models, says attackers hijacked its release pipeline early on Oct 21 and published a malicious version 1.107.0 to PyPI and Docker. LiteLLM says the incident is contained and the bad releases are gone (LiteLLM).

What happened

  • Stolen tokens. LiteLLM says the attacker used three engineers’ GitHub accounts, apparently through stolen GitHub CLI tokens, to change the release pipeline and approve the release. Those tokens are revoked.
  • The window. The PyPI release was live from 11:39 to 13:23 UTC (6:39–8:23 a.m. CT). Docker images tagged v1.107.0, latest, main-latest and main-stable were bad from 11:46 to 14:20 UTC (6:46–9:20 a.m. CT).
  • What it did. JFrog and Chainguard say all seven wheels carried a credential stealer and an hourly remote-code loader that runs whenever Python starts, even if your code never imports LiteLLM. It went after SSH keys, cloud and CI/CD secrets, browser passwords and AI provider keys (JFrog, Chainguard).
  • Hard to spot. The source code on GitHub was clean; the payload was only in the built packages.

What to do

  • Check your installs. If you ran an unpinned pip install or pulled those Docker tags during the windows above, LiteLLM says to assume that machine is compromised.
  • Pin to 1.106.2 or another release from before Oct 21, per LiteLLM.
  • Clean up and rotate. Isolate the machine, remove the malicious startup files before running Python again, and rotate every credential it could reach (Chainguard).

LiteLLM’s private enterprise release wasn’t affected. Chainguard notes this is LiteLLM’s second compromise this year, after a hijacked release in March.

Update, Oct 21 evening: how the attacker got in

LiteLLM’s latest update says it believes it has found the attack chain. It had temporarily widened release permissions on Monday for daily bugfix releases. Of the three stolen tokens, one could run release workflows and one belonged to an owner of LiteLLM’s GitHub organization; the attacker used that one to switch off branch protections, then ran the release workflow from a lookalike repository (Berri-AI/litellm instead of BerriAI/litellm), which signed and published the poisoned packages (LiteLLM).

LiteLLM believes the tokens were taken from the logs of an internal LiteLLM server exposed to the internet, where an unexpected user account turned up. It says it has seen cases where Claude Code puts a user’s GitHub CLI token into a response, which can end up in LiteLLM’s logs when responses are logged. How the attacker got into that server isn’t confirmed; LiteLLM suspects a recently patched vulnerability and says it’s still verifying the details.

Koffee take: pin your versions. Supply-chain attacks hit the tools developers trust most, and a two-hour window was enough to do damage.

Filed Oct 21, 2026 ~4:45 PM CT — Afternoon pass. Updated Oct 21, 2026 ~8:30 PM CT — Evening pass. Sources: LiteLLM — Security incident, v1.107.0 compromised (Oct 21, 2026; updated 20:30 UTC) · JFrog Security Research (Oct 21, 2026) · Chainguard (Oct 21, 2026) · OpenSSF malicious-packages report (Oct 21, 2026). LiteLLM says its investigation is still in progress.