CVE-2026-26326
📋 TL;DR
OpenClaw personal AI assistant versions before 2026.2.14 could expose sensitive configuration secrets to clients with read-only permissions. The vulnerability occurs when the system returns raw resolved configuration values during skill requirement checks, potentially disclosing secrets like Discord tokens. Anyone using OpenClaw with read-scoped clients is affected.
💻 Affected Systems
- OpenClaw
📦 What is this software?
Openclaw by Openclaw
⚠️ Risk & Real-World Impact
Worst Case
Attackers with read-only access could extract all sensitive configuration secrets including API keys, tokens, and credentials, leading to complete account compromise and data breaches.
Likely Case
Unauthorized disclosure of Discord tokens and other API credentials to users or systems with read-only permissions, enabling account takeover and privilege escalation.
If Mitigated
With proper access controls and monitoring, impact is limited to authorized users who already have read access but shouldn't see raw configuration values.
🎯 Exploit Status
Exploitation requires existing read access to the OpenClaw system, making it an information disclosure vulnerability rather than an access control bypass.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: 2026.2.14
Vendor Advisory: https://github.com/openclaw/openclaw/security/advisories/GHSA-8mh7-phf8-xgfm
Restart Required: Yes
Instructions:
1. Update OpenClaw to version 2026.2.14 or later. 2. Restart the OpenClaw service. 3. Rotate any Discord tokens that may have been exposed to read-scoped clients.
🔧 Temporary Workarounds
Restrict read access
allLimit read-scoped client access to only trusted users and systems.
Remove sensitive config from skills
allTemporarily remove or obfuscate sensitive configuration values from skill requirements.
🧯 If You Can't Patch
- Implement strict access controls to limit which users/systems have read permissions
- Monitor and audit all read access to the skills.status endpoint for suspicious activity
🔍 How to Verify
Check if Vulnerable:
Check if OpenClaw version is below 2026.2.14 and if skills.status endpoint returns raw config values to read-scoped clients.Check Version:
Check OpenClaw version in application logs or configuration filesVerify Fix Applied:
After updating to 2026.2.14, verify that skills.status only returns {path, satisfied} objects without raw config values.📡 Detection & Monitoring
Log Indicators:
- Unusual access patterns to skills.status endpoint
- Multiple read requests from single client in short timeframe
Network Indicators:
- Excessive requests to skills.status API endpoint
- Traffic from unauthorized clients to configuration endpoints
SIEM Query:
source="openclaw" AND endpoint="skills.status" AND user_scope="read" AND count > threshold🔗 References
- https://github.com/openclaw/openclaw/commit/d3428053d95eefbe10ecf04f92218ffcba55ae5a
- https://github.com/openclaw/openclaw/commit/ebc68861a61067fc37f9298bded3eec9de0ba783
- https://github.com/openclaw/openclaw/releases/tag/v2026.2.14
- https://github.com/openclaw/openclaw/security/advisories/GHSA-8mh7-phf8-xgfm
