How CVE-2026-25253 exposed every OpenClaw user to RCE — and how to fix it in one command

CVE-2026-25253 (CVSS 8.8) allowed any website to steal OpenClaw auth tokens and achieve remote code execution via a single malicious link, exploiting the excessive agency and lack of semantic firewalls in autonomous AI agents. The vulnerability exemplifies broader architectural gaps—no identity layer, no action authorization, no memory integrity—identified by CrowdStrike, Cisco, and Microsoft. The author proposes Crawdad, a runtime security API independent of the agent that intercepts inbound, outbound, and execution-level actions to prevent prompt injection and privilege abuse.