Executive brief
OpenClaw is a multi-platform AI agent framework used to automate tasks across different operating systems. A flaw in the session management endpoint allowed read-only API callers to terminate active sessions that should have required write permissions, potentially disrupting delegated work and agent operations. The vulnerability only affected HTTP-based identity authentication modes and has been patched.
Technical details
The vulnerability is a missing authorization check (CWE-862, CWE-269) in the session termination endpoint. The POST /sessions/:sessionKey/kill endpoint did not properly enforce write-scoped access control in HTTP identity-bearing authentication modes, allowing operators with read-only credentials to perform a write-class control-plane operation. The fix (commit 54a0878) enforces HTTP operator scopes on session kill requests and gates authorization before session lookup, preventing both unauthorized termination and session existence probing by unauthenticated callers. Exploitation requires network access and valid (read-scoped) identity credentials; no user interaction is needed.
Affected products
- OpenClaw openclaw < 2026.4.2
Timeline
- 2026-04-07: disclosed: Advisory published
- 2026-04-02: patched: Fix committed to main branch
- 2026-04-21: advisory: NVD published CVE-2026-41298