CVE-2026-64622

Network-AI approval reads remained unauthenticated after an earlier fix

Jovancoding · Network-AI · ApprovalInbox

First documented disclosure · Review checkpoint

Summary

Queued command details and approval metadata could be read without the configured secret.

Severity and exploitation evidence

Critical · 9.3/10 · CVSS 4.0. Severity source.

Exploitation not established: Reviewed public sources describe the weakness, a correction or a controlled demonstration; 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

Only mutating approval routes applied the authorization check; read and event-stream routes retained wildcard CORS.

Conditions required

The ApprovalInbox HTTP interface is reachable, including from a visited web page.

Potential impact

Disclosure of command text, paths, justifications and risk labels.

Affected versions and configurations

5.12.2–5.13.3.

Fix and mitigation status

5.13.4.

Practical next steps

Upgrade and verify authorization/CORS on list, detail, statistics and stream endpoints.

Operating lesson

Read access to approval queues can be sensitive even when approval actions are protected.

Evidence and limits

Historical backfill; original PoC demonstrates unauthorized reads, not executed malicious approvals. This is an incomplete earlier remediation. CNA publication is 2026-07-20; this is distinct from the earliest reviewed public source and correction dates. VulnCheck assessments: CVSS 4.0 9.3; CVSS 3.1 7.5.

Sources