1. Executive summary
OpenClaw versions prior to 2026.5.18 contain an authorization bypass vulnerability (CVE-2026-53821) in the Gateway WebSocket control plane. The Gateway accepts client-declared operator scopes before binding them to a server-approved pairing flow or a trusted-proxy authorization baseline, allowing unpaired or restricted trusted-proxy Control UI clients to obtain cached operator.admin authority on live WebSocket connections and execute admin-gated Gateway RPCs. The publishing GitHub Advisory classifies the issue as High severity; the CVE is currently marked Unreviewed by NVD and no CVSS vector has been resolved. EMEA financial services entities running OpenClaw — including as a third-party component in trading, ops, or automation pipelines — should treat this as a P1 patch item pending further review.
2. Regulatory framing
| Article | Trigger | Practical impact |
|---|---|---|
| DORA Art. 28 | OpenClaw is a third-party ICT software component embedded in ICT systems supporting business functions | Financial entities must include OpenClaw in their ICT third-party risk register and assess whether the vulnerability triggers notification, patch SLA, or exit-clause obligations with the providing vendor/integrator |
| NIS2 Art. 21(2)(d) | OpenClaw is a supply-chain software dependency | In-scope entities must evaluate the OpenClaw dependency as part of supply-chain security measures and ensure vulnerabilities in third-party components are identified and addressed |
3. Technical analysis & attack chain
The vulnerability is in the OpenClaw Gateway's WebSocket authorization logic. The Gateway accepts operator scope claims declared by the WebSocket client before validating them against either:
- A completed device-pairing / bootstrap flow, or
- A trusted-proxy authorization baseline.
This means an attacker who can reach the WebSocket endpoint — including via an unpaired Control UI client or a restricted trusted-proxy client — can claim operator.admin scope and the Gateway will honour it on the live connection, granting access to admin-gated Gateway RPCs.
Attack chain (confirmed)
- Reach the WebSocket endpoint. Attacker reaches the OpenClaw Gateway WebSocket listener (typically the Control UI port) directly, or via a trusted-proxy that does not enforce pairing.
- Declare
operator.adminscope. Attacker opens a WebSocket session and declaresoperator.adminscope in the client handshake before any pairing or trusted-proxy authorization has completed. - Gateway honours the claim. Because the Gateway binds to the client-declared scope before validating against the pairing/proxy baseline, the session is granted
operator.adminauthority. - Execute admin-gated Gateway RPCs. Attacker invokes admin-gated RPCs against the live WebSocket session, obtaining the same authority as a fully paired operator.
Technical specifics
- Affected versions: OpenClaw prior to 2026.5.18
- Fixed in: OpenClaw 2026.5.18
- Component: Gateway WebSocket control plane
- Affected scope:
operator.admin(admin-gated Gateway RPCs) - Trigger conditions: Unpaired Control UI clients OR restricted trusted-proxy clients
- Authorization model bypassed: Server-approved pairing OR trusted-proxy authorization baseline
- Related vulnerabilities in OpenClaw (context):
- GHSA-ffhm-8fwq-7q27 (corpus-1): OpenClaw before 2026.5.12 — allowlist bypass in PowerShell encoded-command handling via abbreviated flag aliases not recognized by the allowlist parser. Remote authenticated operators can bypass execution allowlist checks to execute arbitrary PowerShell content.
- GHSA-jj27-4rc8-m6mm (corpus-2): OpenClaw before 2026.5.4 — authorization bypass in the bundled device-pair plugin allowing non-owner authorized chat senders to issue device-pairing bootstrap codes without proper scope validation, granting persistent credentials until manual removal.
Caveats
The CVE is currently marked "Unreviewed" by NVD; no CVSS vector has been resolved. The "High severity" classification is the GitHub Advisory's own assessment. No public exploitation, threat-actor attribution, or in-the-wild campaign has been confirmed in the source material. The two related corpus advisories indicate OpenClaw has had a pattern of authorization-scope and allowlist-parser weaknesses across recent releases (2026.5.4, 2026.5.12, 2026.5.18), suggesting the project's authorization model has not been consistently hardened.
4. Mitigation & containment
P1 — within 24 hours
- Patch to 2026.5.18 or later. Upgrade all OpenClaw deployments (Gateway, Control UI clients, bundled plugins) to OpenClaw 2026.5.18 or the latest available release.
- Inventory exposure. Identify all OpenClaw instances, WebSocket Gateway endpoints, and trusted-proxy frontends in the estate. Confirm whether any are internet-reachable.
- Restrict WebSocket reachability. Until patched, restrict the OpenClaw Gateway WebSocket listener to known operator IPs / VPN ranges at the network layer. Firewall or disable any internet-facing Control UI frontends.
P2 — within 72 hours
- Audit
operator.adminactivity. Review Gateway RPC logs and WebSocket session logs for the period since the last known-good state. Look for sessions that invoked admin-gated RPCs without a corresponding completed pairing event or trusted-proxy authorization record. - Revoke cached credentials. Force re-pairing of all Control UI clients and rotate any cached operator credentials issued prior to the patch.
- Validate trusted-proxy configuration. Confirm trusted-proxy frontends enforce the authorization baseline (not just pass-through) and reject sessions that have not completed pairing.
P3 — within 7 days
- Review related OpenClaw advisories. Assess exposure to GHSA-ffhm-8fwq-7q27 (PowerShell encoded-command allowlist bypass, fixed in 2026.5.12) and GHSA-jj27-4rc8-m6mm (device-pair plugin authorization bypass, fixed in 2026.5.4). Patch to a release that includes all three fixes.
- Vendor / contractual review. Where OpenClaw is supplied under a third-party contract, invoke DORA Art. 30 patch/SLA provisions and confirm the supplier's vulnerability-disclosure and incident-cooperation obligations.
- Concentration-risk assessment. If OpenClaw is used pervasively across business functions, evaluate whether DORA Art. 29 concentration-risk assessment is warranted.
5. Indicators of compromise
No indicators of compromise available in the source material.
6. Detection
title: OpenClaw Gateway WebSocket operator.admin Scope Claim Before Pairing
id: 9a8c7b3e-2d4f-4e1a-b5c6-7d8e9f0a1b2c
status: experimental
description: Detects WebSocket sessions to OpenClaw Gateway that declare operator.admin scope before pairing or trusted-proxy authorization has completed.
author: Adverse Trace
date: 2026-06-13
references:
- https://github.com/advisories/GHSA-vcg8-6gf2-cg9j
logsource:
product: openclaw_gateway
category: websocket_session
detection:
selection_scope:
websocket_scope_claimed: "operator.admin"
selection_unauth:
pairing_completed: false
trusted_proxy_authorized: false
condition: selection_scope and selection_unauth
level: high
tags:
- attack.initial_access
- attack.t1078
7. Sources
- GitHub Security Advisories — GHSA-vcg8-6gf2-cg9j: OpenClaw before 2026.5.18 accepts WebSocket client-declared operator scopes before binding to server-approved pairing or trusted-proxy authorization baseline. https://github.com/advisories/GHSA-vcg8-6gf2-cg9j (published 2026-06-13)
- GitHub Security Advisories — GHSA-ffhm-8fwq-7q27: OpenClaw before 2026.5.12 contains an allowlist bypass vulnerability in PowerShell encoded-command handling. https://github.com/advisories/GHSA-ffhm-8fwq-7q27
- GitHub Security Advisories — GHSA-jj27-4rc8-m6mm: OpenClaw before 2026.5.4 contains an authorization bypass vulnerability in the bundled device-pair plugin. https://github.com/advisories/GHSA-jj27-4rc8-m6mm
8. Adverse Trace position
Severity: High (per publishing GitHub Advisory; CVE currently unreviewed by NVD, no CVSS vector resolved). Client impact: any EMEA financial services entity running OpenClaw prior to 2026.5.18 — particularly where the Gateway WebSocket is reachable from untrusted networks or via trusted-proxy frontends without enforced pairing — is exposed to admin-gated RPC execution by unauthenticated or restricted clients. Next steps: Adverse Trace will (a) monitor NVD for the resolved CVSS vector and update this advisory, (b) track for any public PoC or in-the-wild exploitation, and (c) re-assess if a confirmed threat-actor attribution emerges.
Published via PulseTrace — Adverse Trace threat intelligence.