A missing scope check in OpenClaw's /pair approve workflow allowed any user with basic pairing permissions to silently grant themselves admin access, enabling full instance takeover without audit trail.
CVE-2026-33579 exposes a critical privilege-escalation flaw in OpenClaw's device-pairing workflow. When a user invoked /pair approve, the server failed to validate whether the requester's scope covered the target pairing's requested privileges — it only checked that the requester held *some* pairing scope. Any operator could therefore approve admin-level pairings for their own attacker-controlled devices, escalating to full instance control with no audit entry. Because pairings don't require an MFA challenge, exploitation was a single HTTP request.
The Actual Culprit
The /pair approve endpoint used a truthy check on the requester's pairing scope instead of validating that the scope covered the requested privilege level. Audit logging for pairing decisions was also absent.
Checking that a caller has *some* privilege is not the same as checking they have *the required* privilege. Every authorization check should be subset-based, not existence-based.
When approval decisions don't produce audit rows, you cannot retroactively identify which pairings were exploitative. Logging is a security control, not a debugging convenience.
Loading comments...