Executive brief
OpenClaw, a tool used for managing AI model interactions, contains a security flaw in how it handles specific model override requests. An attacker with basic user access could bypass administrative security policies to perform restricted actions or access sensitive data. This could lead to unauthorized use of AI resources or exposure of internal configuration details.
Technical details
An authorization bypass vulnerability exists in OpenClaw's OpenAI-compatible HTTP model override feature. The flaw stems from missing or incorrect authorization checks (CWE-862/CWE-863) when processing specific input paths, allowing a low-privileged authenticated user to circumvent administrative policies. By exploiting misconfigured input paths, an attacker can execute restricted operations that should be reserved for higher-trust operators. The vulnerability is reachable over the network if the affected feature is enabled. A fix is available in version 2026.6.8; users are advised to upgrade or restrict the feature to trusted operators as a mitigation.
Affected products
- OpenClaw OpenClaw < 2026.6.8
Timeline
- 2026-06-30: advisory: GitHub Security Advisory published
- 2026-07-13: disclosed: NVD publication date