A critical vulnerability in OpenClaw, a rapidly adopted open source AI agent tool, allowed malicious websites to hijack a developer's local AI agent via WebSocket connections to localhost without any user interaction. The flaw exploited OpenClaw's implicit trust of localhost connections and lack of brute-force protections on
•5m read time• From darkreading.com
Table of contents
High Severity VulnerabilityA Growing List of Security IssuesThe Need for a New ApproachSort: