Disclosed weakness · no CVE recorded

OpenClaw could start provider discovery after model authorization was revoked

OpenClaw · OpenClaw agent runtime · model selection

First documented disclosure · Review checkpoint

Summary

An agent selecting a model outside the catalog could start provider discovery after its authorization for that selection had expired or been revoked.

Severity and exploitation evidence

No CVSS score recorded in this review.

Exploitation not established: The project documented the defect in development testing. The cited sources identify no production exploitation.

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 runtime detected the loss of authority after provider preparation had already started.

Conditions required

Model selection involved an off-catalog provider and its captured authorization had been revoked before preparation.

Potential impact

A provider discovery call could occur after revocation. The existing final check still rejected the model selection.

Affected versions and configurations

The upstream reports document affected behavior without a complete range of versions.

Fix and mitigation status

OpenClaw 2026.11.4 lists authorization checks around off-catalog model discovery.

Practical next steps

Use 2026.11.4 or a later release containing the authorization check. Review provider audit activity when investigating a revoked model selection.

Operating lesson

Check current authorization before calling a provider and again before accepting its result.

Evidence and limits

October 20 is the issue-specific upstream fix date. The maintainer tested with a synthetic provider. The sources do not establish model inference, a committed configuration change or access to customer data.

Sources