Disclosed weakness · no CVE recorded
LiteLLM tool deny rules could miss Anthropic and Responses requests
LiteLLM · LiteLLM · Tool Permission guardrail
First documented disclosure · Review checkpoint
Summary
Configured pre-call tool deny rules did not fire on the tool shapes used by the Anthropic Messages and Responses APIs.
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.
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 guardrail read Chat Completions tool names without interpreting the other endpoint formats and namespace members.
Conditions required
The operator enables Tool Permission rules and accepts requests through the affected API formats.
Potential impact
Disallowed tools could reach the provider despite the configured tool policy.
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; a full earlier affected range is unverified.
Practical next steps
Upgrade to 1.110.0 and validate intended deny and rewrite rules against each API format used by clients.
Operating lesson
Test policy enforcement across every supported request representation.
Evidence and limits
The public PR dates to September 13 and merged November 4. Tagged verification: the pre-call guardrail interprets endpoint-specific tool fields, hosted tools and namespace members before block or rewrite decisions.
Sources
- fix(guardrails)!: check Anthropic and Responses request tools in tool_permission pre_call
Primary source ·
- v1.110.0 release notes
Primary source ·
- LiteLLM 1.110.0 tool policy enforcement
Primary source ·