The Identity-to-Execution Gap
Primary Barrier to Confident AI Agent Adoption in Financial Services
Archived context
This page preserves a March 2026 standards submission. It is historical context, not a current production-authority claim.
The current public claim boundary is governed by Boundaries and Non-Claims, the sample receipt, and Public Proof Desk. Release-summary detail belongs in static zlar.ai proof artifacts. This page is historical context and does not claim production authority, enterprise readiness, public external attestation, sovereign recognition, or coverage of unrouted surfaces.
On March 29, 2026, ZLAR founder Vincent Nijjar submitted this paper to the NIST Center for AI Standards and Innovation (CAISI) as part of their Listening Sessions on Sector-Specific Barriers to AI Adoption, focused on financial services.
The submission identifies a single structural gap preventing confident AI agent deployment in regulated financial services: no standard governs what agents do after authentication. The credential boundary is addressed. The execution boundary is not.
The paper presents evidence from the Cloud Security Alliance, the FIFAI II Report, Gartner, and Rubrik Zero Labs quantifying the gap, maps the compounding regulatory friction across SOX, SR 11-7, OCC 2023-17, and BCBS 239, and presents ZLAR as a working execution-boundary governance system with prior Apache-2.0 public distribution. Current public claims remain bounded by the v3.4.0 release and do not assert production authority, live approval-channel delivery or health, a live records system, a production records adapter or service, persistent runtime profile installation or activation, exactly-once effect semantics, production-grade anti-rollback protection, stale-lock recovery, multi-host coordination, production-grade durable storage, or tamper resistance for the consumed-receipt store, direct filesystem side-door closure, host process side-door closure, or live machine coverage.