Threat-Model the Egress, Not the Model
A developer argues that AI features should be secured by threat-modeling the outbound payload rather than the model itself, since every prompt is a data transfer. The proposed approach is a "boundary ledger" and a delibe…