AI Briefing
KO

CVE Alert Bomb

·2026.04.16 23:31

Key point

Among 138 CVEs in the OpenClaw ecosystem, the entries critical for self-hosting were singled out as warnings.

Details

jgamblin's GitHub repository OpenClawCVEs tracks CVEs across OpenClaw, clawdbot, and moltbot, with 138 cumulatively documented between February and April. 13 were added in April alone, with an average CVSS of around 7.0.

The key vulnerabilities requiring immediate updates are as follows.

  • CVE-2026-25253 (CVSS 8.8): One-click RCE via authentication token theft. Patched in 2026.1.29, with prior versions vulnerable to malicious websites.
  • CVE-2026-33579 (CVSS 9.8): A critical vulnerability patched on 2026.3.28, leaving self-hosted environments exposed for several days before disclosure on April 1.
  • CVE-2026-35639 (CVSS 8.7): Scope validation bypass in the device.pair.approve handler. Caused by failure to verify the request token's permissions, fixed on 2026.4.5.
  • CVE-2026-35641 (CVSS 8.4): .npmrc credential leak during plugin installation. If write access existed within a directory scanned by OpenClaw, package resolution could be redirected to a malicious registry; patched on 2026.4.5.
  • CVE-2026-34510: A Windows path traversal vulnerability. Can expose sensitive files outside the working directory, particularly dangerous when run with administrator privileges.

On the operational side, it's emphasized to bind the gateway to 127.0.0.1 instead of 0.0.0.0. According to Blink's investigation, 63% of 135,000 publicly exposed OpenClaw instances lack authentication, meaning the April vulnerabilities are remotely exploitable without credentials in such environments.

Additional recommended countermeasures include setting a 64-character strong gateway token, blocking port 18789, and using a VPN like Tailscale instead of opening ports for remote access. For those using managed hosts, it's advised to first check whether auto-patching is enabled and the CVE response time.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.