The ClawHavoc campaign exposed a fundamental architectural flaw in OpenClaw: all skills share a single plaintext credential file with no delegation model. Over 800 malicious skills on ClawHub exploited this to exfiltrate tokens. The real fix isn't better secret hygiene — it's applying OAuth-style delegation to AI agents. Auth0

8m read timeFrom auth0.com
Post cover image
Table of contents
What We Mean by DelegationWhat ClawHavoc Looked Like from Where We SitA Tool You Can Use Right NowGoing Deeper with Fine-Grained AuthorizationThe Pattern Repeats

Sort: