Tools AI Risk Radar ai-incident-0103
Incident record
OpenClaw let other chat members approve Claude Code actions meant for its owner
- Severity
- High
- Status
- Patched
- Type
- Agent Hijack
- Target
- OpenClaw (npm openclaw) before 2026.7.1, the MCP channel bridge for Claude Code permission prompts
- Actor
- researcher
- CVE
- CVE-2026-100585
What happened
OpenClaw relays Claude Code's permission prompts into chat channels through its MCP channel bridge, and before 2026.7.1 it let any authorised sender with channel command access answer a prompt meant for the channel owner. A non-owner could approve or deny the pending request, so a Claude Code action could proceed without the owner's consent. Scored CVSS 3.1 8.0; the impact depends on the pending action and the host capabilities the Claude Code run requested.
The fix, OpenClaw 2026.7.1, shipped on npm on 13 July 2026; the CVE record reached NVD on 26 September. The flaw sits in OpenClaw's bridge, the third-party layer between the chat channel and Claude Code's own permission prompt.
Sources
- NVD record CVE-2026-100585nvd.nist.gov/vuln/detail/CVE-2026-100585
- OpenClaw advisory GHSA-p5g8-m35v-7m82github.com/openclaw/openclaw/security/advisories/GHSA-p5g8-m…
One record from the AI Risk Radar, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 13 October 2026