Disclosed weakness · no CVE recorded

LiteLLM could retain model content in logs despite message redaction

LiteLLM · LiteLLM · chat completion logging

First documented disclosure · Review checkpoint

Summary

Logs could retain reasoning, citations, extracted file text and search results after message redaction was enabled.

Severity and exploitation evidence

No CVSS score recorded in this review.

Exploitation not established: The reviewed sources describe a correction or controlled reproduction; malicious exploitation against users 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

The redaction path omitted sensitive values nested under provider_specific_fields.

Conditions required

Message redaction is enabled and the provider returns content in the affected auxiliary fields.

Potential impact

People or systems with access to those logs could see content the operator intended to redact.

Affected versions and configurations

The reviewed upstream report does not establish a complete affected-release range.

Fix and mitigation status

1.110.0 contains the correction; other release lines were not independently established.

Practical next steps

Upgrade to 1.110.0 and review retained logs for affected content according to the operator’s retention policy.

Operating lesson

A redaction policy must cover auxiliary provider fields as well as the visible message.

Evidence and limits

The maintainer deliberately retains unknown metadata keys, and top-level fields outside provider_specific_fields are outside this patch. Tagged verification: selected reasoning fields are replaced and content-bearing provider result records are nulled.

Sources