Yeah we discussed this back in August about comparing concrete revocation scenarios. (My user name was under Nash then)
This is something we’ve since tested internally. An agent can understand the person correctly and still act on permission that has expired.
We built a small offline prototype to test exactly that. In simulated scenarios, permission was revoked or expired after planning. When the system tried to proceed, it detected the change and stopped before contacting an external service.
It seems that would be relevant for Holon. Understanding should carry forward, but yesterday’s permission shouldn’t automatically carry with it. Permission would need to reflect the present.
I’m not sure where you all are at but I think there are beneficial exploratory lanes here.