Oso

Oso is an authorization platform for applications, which has moved its emphasis toward securing AI agents. The product provides the permission model an application uses to decide who may do what, alongside an audit trail of agent sessions and alerting on activity that falls outside expected behaviour.

The pivot toward agents follows a real change in how software acts. When a person uses an application, their permissions bound what happens. When an agent acts on their behalf it may touch many systems in one session, and the interesting question stops being whether a single request is allowed and becomes whether the pattern of a whole session was reasonable. A full audit trail of agent sessions answers a question that per request checks cannot.

The vendor cites research that employees leave the large majority of their granted permissions unused, and observes that agents will not. That is a sharp way of putting a genuine risk: over granting has historically been safe because people did not exercise the access they held, and an agent working through a task will exercise all of it.

The platform retains its application authorization foundations, with a modelling approach and libraries for expressing permissions, alongside education material through its Authorization Academy.