Disclosed weakness · no CVE recorded

LiteLLM streaming fallback could repeat a tool action

LiteLLM · LiteLLM Router · Responses API streaming fallback

First documented disclosure · Review checkpoint

Summary

A client could execute the same tool action twice when a provider stream failed after delivering the first call.

Severity and exploitation evidence

No CVSS score recorded in this review.

Exploitation not established: The reporter demonstrated duplicate execution with a real client in a controlled test. The reviewed evidence identifies no hostile use.

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 router tracked delivered text when deciding whether to retry, but failed to treat a delivered tool call as output.

Conditions required

A streaming Responses client executes tool calls, and a retryable provider error occurs after delivery with a fallback available.

Potential impact

A single request could cause two external changes. The effect depends on the tool and whether repeated execution is safe.

Affected versions and configurations

The reporter confirmed the behavior in 1.102.1 and on main. The full affected range is unspecified.

Fix and mitigation status

LiteLLM lists the fix in its development release 1.111.0-dev.4.

Practical next steps

Deploy a build containing the fallback correction. Inspect repeated tool records for any action that could have changed external state twice.

Operating lesson

Track tool-call delivery independently of text before retrying a model response.

Evidence and limits

The maintainer documents a correction. The cited sources provide no issue-specific severity rating or CVE.

Sources