Multiple OpenClaw Vulnerabilities Enables Policy Bypass and Host Override

Researchers disclosed three moderate-severity vulnerabilities in the OpenClaw AI agent framework (formerly Clawdbot/Moltbot) distributed via npm: a gateway configuration mutation that lets a model persistently alter operator-trusted settings, a bundled-tool policy-enforcement bypass that can activate tools after filtering, and a workspace host override that can redirect credentialed requests to attacker-controlled servers. The development team patched the issues in OpenClaw 2026.4.20 and recommends immediate upgrades to prevent credential exposure and local policy compromise.