We audited authorization in 30 AI agent frameworks — 93% rely on unscoped API keys https://www.reddit.com/r/netsec/comments/1ruefpo/we_audit…