The SVRNOS Non-Content Safety Attestation

The Non-Content Safety Attestation (NCSA) specification is part of the SVRNOS governance research published by Sushee Nzeutem (Sovereign OS LLC). It is the evidentiary receipt format that complements the SVRNOS Governance Error Register (GER) classification taxonomy and the SVRNOS 7-Layer Model architectural map: GER names what kind of governance failure occurred, the 7-Layer Model identifies where in the technical stack the failure or control sits, and NCSA provides the signed, cryptographically verifiable receipt that the governance layer ran at all. Abstract Private AI architectures built on Trusted Execution Environments (TEEs) ship a strong privacy guarantee: conversation content does not leave the enclave. They do not, on their own, ship a verifiable safety guarantee: that a governance layer ran, what risk category it detected, what action it took, and who was accountable when it did. NCSA v0.2 specifies a format for a non-content safety attestation: a signed, cryptographically verifiable, procedural statement that an in-enclave governance layer evaluated a session and produced a defined outcome, with no transcript or content recoverable from the attestation itself. v0.2 makes four contributions on top of v0.1. First, an explicit design discipline: the signed payload is procedural-only, not a record of reasoning or judgment. Second, payload field additions that surface governance distinctions v0.1 conflated: threshold-per-action, accountable-party identity, constraint locus, temporal recognition lag, admissible-alternative space, and a reference to the SVRNOS Governance Error Register (GER) taxonomy. Third, usage rules covering re-admissibility when an NCSA receipt crosses a system boundary or is retrieved from persistence. Fourth, an explicit positioning statement vs the runtime governance instrumentation projects emerging in mid-2026 (Zenity-led Agent Control Standard and Microsoft Agent Control Specification), with which NCSA composes rather than competes. v0.2 is published for public comment before v1.0. Comments are invited from implementers, auditors, regulators, researchers, platform operators, and the contributors named in the Acknowledgments. NCSA in one sentence: NCSA is the signed receipt that proves the governance layer ran. It is not the safety layer itself, the audit dashboard, or the explanation record. It is the procedural evidence that the layer existed and fired. Previously published as v0.1 at https://svrnos.com/research/non-content-safety-attestation on May 14, 2026. This v0.2 deposit on Zenodo introduces the four contributions named above. Companion publications SVRNOS 7-Layer Model of AI Governance v0.1 — https://doi.org/10.5281/zenodo.20600670 (the architectural locator map) SVRNOS Governance Error Register (GER) v0.2 — DOI pending same-day publish (the failure-classification taxonomy) Full methodology, in-progress work, regulatory engagement notes, and contributor acknowledgments at https://svrnos.com/research/non-content-safety-attestation.