AI security is becoming identity security
TechCrunch AI
31 July 2026
How this applies to you
If you are testing AI agents with access to files, inboxes, CRMs or cloud tools, do not give them broad access by default. Give each agent the smallest permission set it needs and review what it touched.
Okta is buying Permiso to add identity threat detection as companies secure AI agents and other non-human identities. This points to a quiet shift: the more agents act inside business systems, the more permissions matter.
Why it matters. If you are testing AI agents with access to files, inboxes, CRMs or cloud tools, do not give them broad access by default. Give each agent the smallest permission set it needs and review what it touched.
