A Policy Algebra for Trust-Preserving Agentic AI Execution

Excerpt

arXiv:2608.16402v1 Announce Type: new Abstract: Large language model-based agentic frameworks primarily optimize capability: whether an agent can reason, retrieve information, call tools, delegate work, and complete a goal. Enterprise execution requires a stronger property. A successful result is not reliable if it was produced through unauthorized data access, widened delegated authority, unapproved side effects, unrecoverable budget consumption, or incomplete evidence. This paper defines relia