critical severityConfidence highHistorical record

LiteLLM Enterprise proxy leaks its master key to a request with no API key (GHSA-w4c7-34wj-hqm8)

Direct answer

LiteLLM Enterprise proxy leaks its master key to a request with no API key (GHSA-w4c7-34wj-hqm8), also tracked as GHSA-w4c7-34wj-hqm8 / LiteLLM master key leak / LiteLLM Enterprise unauthenticated file read is a critical-severity threat campaign HOL Guard tracked from Oct 27, 2026 to Oct 27, 2026, verified against 5 primary sources and last reviewed Oct 27, 2026. Active tracking has ended and the record remains published as a permanent historical research record.

An unauthenticated request to any LLM route on a LiteLLM proxy running with an Enterprise license could make the proxy read a local file and send its contents to a URL chosen by the attacker. Reading the process environment exposes LITELLM_MASTER_KEY and provider API keys, which gives full admin on the proxy; in some configurations this leads to arbitrary Python execution as root. The root cause, per PR #48599, is that failure logs for rejected requests were built from the raw request body, so a caller with no valid key could set per-request logging parameters such as the Langfuse or LangSmith host and the GCS bucket and service-account path. Fixed in litellm 1.103.10, 1.104.8, 1.105.6 and 1.106.5, whose release notes describe the change only as a logging change with 'No action needed to upgrade'. Open source deployments without a license are not affected.

Also tracked as GHSA-w4c7-34wj-hqm8 · LiteLLM master key leak · LiteLLM Enterprise unauthenticated file read · litellm 1.106.5 security fix · LiteLLM failure log allowlist PR 48599

First observed
Oct 27, 2026
Last observed
Oct 27, 2026
Last reviewed
Oct 27, 2026
Tracking ended
Oct 27, 2026

Record exports

JSON recordArtifact CSVSTIX 2.1 bundle

Uncertainty and limitations

Read the evidence limits before acting on this record.

BerriAI has not published a full exploit chain; the link between the caller-controlled logging fields and the file read plus exfiltration is HOL's reading of PR #48599 and the advisory text. No CVE ID assigned yet. The advisory reports no exploitation and the issue is not on CISA KEV. The number of Enterprise-licensed proxies exposed is unknown.

  • Open source LiteLLM proxies without an Enterprise license are not affected; the vulnerable code paths only load with an active license.
  • Applications that use the litellm Python SDK without running the proxy server are not in scope of the advisory.
  • Proxies already on 1.103.10, 1.104.8, 1.105.6 or 1.106.5 are fixed.
  • This is not a supply-chain compromise like 1.107.0; the published packages were not tampered with.
  • No exploitation has been reported and the issue is not on CISA KEV as of publication.

Guard coverage and policy

What Guard's reviewed assertions do and do not claim for this campaign, and the reviewed starting points for defense.

No Guard coverage mapping or campaign-specific policy guidance is published for this record.

Safe next step. Campaign guidance is defensive context. It does not auto-apply policy or claim universal campaign coverage.

Open policy controlsCampaign-specific emergency policy starting point

Timeline

Reviewed events, oldest first, each tied to a verified source.

  1. Vendor action

    Fixed litellm 1.104.8, 1.103.10 and 1.105.6 uploaded to PyPI between 19:39 and 19:42 UTC.

    Source:PyPI litellm release history

  2. Vendor action

    PR #48599 merged: failure logs for rejected requests built from an allowlist; GCS logger pinned to its configured bucket and service account.

    Source:BerriAI/litellm PR #48599: build proxy-only failure logs from an allowlist

  3. Vendor action

    GitHub releases v1.103.10, v1.104.8, v1.105.6 and v1.106.5 published, labeled as logging changes with no upgrade action needed.

    Source:LiteLLM v1.106.5 release

  4. Disclosure

    BerriAI publishes GHSA-w4c7-34wj-hqm8, critical, CVSS 10.0, crediting Dylan Ong (Karmasec) and Cyrill Bannwart (Compass Security).

    Source:GHSA-w4c7-34wj-hqm8: Unauthenticated file read leaks master key on Enterprise proxy

  5. Update

    HOL Guard publishes operator analysis with version and license check plus rotation steps.

    Source:HOL Guard analysis: LiteLLM Enterprise proxy master key leak

Publication clock: 0.3 hours from reviewed disclosure timestamp to HOL publication; 12-hour critical target; within target.

Reviewed artifacts

The specific artifacts this record concerns, as reviewed. Names and versions are shown as text; a registry link is not recorded here.

  • litellm@< 1.103.10; 1.104.x < 1.104.8; 1.105.x < 1.105.6; 1.106.x < 1.106.5 (Enterprise license only)pypipackage
  • ghcr.io/berriai/litellm@fixed: v1.106.5, v1.105.6, v1.104.8, v1.103.10dockerpackage

Sources

Every claim on this record is traceable to the sources below.

Campaign questions, answered

An unauthenticated request to any LLM route on a LiteLLM proxy running with an Enterprise license could make the proxy read a local file and send its contents to a URL chosen by the attacker. Reading the process environment exposes LITELLM_MASTER_KEY and provider API keys, which gives full admin on the proxy; in some configurations this leads to arbitrary Python execution as root. HOL Guard tracked the campaign from Oct 27, 2026 to Oct 27, 2026.

No. Active tracking ended Oct 27, 2026; this page remains as a labeled historical research record and does not imply the threat is currently active.

The record cites 5 primary sources, names 2 reviewed artifacts, and was last fully reviewed Oct 27, 2026 by HOL Guard Security Publishing. Its uncertainty and limitations are stated on the record itself.

Campaign guidance is defensive context. It does not auto-apply policy or claim universal campaign coverage.

The permalink above is permanent. The full record can be downloaded from this page as JSON, an artifact CSV, and a STIX 2.1 bundle, and corrections can be requested through the published correction path on the record. BerriAI has not published a full exploit chain; the link between the caller-controlled logging fields and the file read plus exfiltration is HOL's reading of PR #48599 and the advisory text. No CVE ID assigned yet.

Reviewed in full by HOL Guard Security Publishing.

Published
Last full review
Last modified

Report a correction

Record HGTC-2026-424CAF68