Trapdex Subscribe

CampaignsAI plugin / MCP / skill

TeamPCP backdoors the LiteLLM AI gateway SDK on PyPI (1.82.7 and 1.82.8)

Do this now

  1. Search environments for those versions and for litellm_init.pth in site-packages
  2. Rotate every credential reachable from affected hosts and CI, Kubernetes included
  3. Block the listed domains and IP

Indicators to check

Look for any of these on your machines or in your logs:

  • packagelitellm==1.82.7
  • packagelitellm==1.82.8

CSVJSONAll 5 indicators

What happened

With a PyPI token stolen through the earlier Trivy compromise, the attacker published two LiteLLM versions carrying a .pth file that runs a credential stealer whenever Python starts. It targeted cloud, SSH, Kubernetes, LLM-provider and wallet secrets. PyPI quarantined the package after about three hours.

  • PyPI
  • Python
  • Kubernetes

Attack chain

  1. Lure

    Trusted update of a widely used LLM proxy and SDK

    First seen

  2. Artifact

    litellm 1.82.7 and 1.82.8 with litellm_init.pth

  3. Actor

    TeamPCP (attribution: OpenSourceMalware and others; basis: the actor's own defacement messages, the tpcp exfil archive name, and the direct chain from the Trivy compromise)

  4. Infrastructure

    • Lookalike LiteLLM exfil domain
    • Secondary C2

    Last seen

Indicators

CSVJSON

All values are defanged (how to read them).

TypeValueCopy
packagelitellm==1.82.7
packagelitellm==1.82.8
domainmodels[.]litellm[.]cloud
ip46[.]151[.]182[.]203
domaincheckmarx[.]zone

Confidence: Moderate

Strong evidence, not yet confirmed by our own analysis. See the confidence scale for what each level requires. The sources below are the evidence.

Sources

Notes

  • Downstream of trivy-teampcp-action-tag-hijack (publisher-pipeline). Separate from the Oct 2026 LiteLLM 1.107.0 incident.

Corrections

No corrections.

See the corrections policy. Spotted an error? Tell us.