Disclosed weakness · no CVE recorded

LiteLLM rejected requests could still exhaust proxy memory

LiteLLM · LiteLLM Proxy · authentication and rate limiting

First documented disclosure · Review checkpoint

Summary

Repeated large requests from blocked or over-limit keys could consume memory before being refused.

Severity and exploitation evidence

No CVSS score recorded in this review.

Exploitation not established: The maintainer reports a customer retry-related outage and controlled reproduction. Malicious exploitation is not established.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Authentication buffered the body before deciding that the key or header-evaluable limits already required rejection.

Conditions required

The client can submit large bodies to metered endpoints using a blocked or rate-limited key and repeats requests.

Potential impact

Proxy processes can run out of memory and interrupt other users’ requests.

Affected versions and configurations

The maintainer names 1.108.4 in the customer report; a complete affected-release range is unverified.

Fix and mitigation status

1.109.2 contains the correction and its follow-up adjustments.

Practical next steps

Update the proxy, limit request-body size and enforce retry backoff. Verify edge limits for paths that cannot be rejected from headers.

Operating lesson

Reject known-denied traffic before allocating resources for its contents.

Evidence and limits

The November 12 PR reports a customer retry-related outage on 1.108.4 and controlled memory measurements. Tagged 1.109.2 calls the early rejection helper before body parsing. The correction is cache-dependent; body-dependent limits and cache misses retain the normal path. Later fixes preserve bounded logging and use current cached team limits.

Sources