Abstract
This document discloses, with sufficient detail to enable a person of ordinary skill in the art to build it, an egress data-loss-prevention (DLP) policy engine that decides what outbound text an autonomous AI system is permitted to transmit to third-party language-model providers. The engine's distinguishing characteristic is that each policy is scoped jointly by three orthogonal dimensions — the internal pipeline route, the identity of the originating AI persona, and the human user role — and that a single first-match decision returns both a content action (allow, redact, block) and the provider that may receive the resulting text. Policies are stored in a relational database, evaluated in ascending priority order, and backed by a fail-closed lifecycle: an enforcing default policy set is seeded when storage is empty, an in-memory default set enforces during database outage, and a single-active-policy invariant is preserved by a partial unique index. Redaction is performed by descending-offset span splicing. This disclosure is published to establish dated public prior art and to bar patent claims over the disclosed mechanism.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Assuncao, gustavo matthew, "Fail-Closed, Persona-Scoped DLP Egress Policy Engine with Classification-Conditional Provider Steering", Technical Disclosure Commons, (July 13, 2026)
https://www.tdcommons.org/dpubs_series/10869