Published June 18, 2026 | Version 1.1

Amagi Open Specification v1.1: The Standard for Hardware-Enforced AI Safety

  • 1. AGIAM Technologies Ltd

Contributors

Description

License: This work is dual-licensed.
(1) Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) for non-commercial use — https://creativecommons.org/licenses/by-nc/4.0/legalcode.
(2) S-APL v2.0 (Strict Architectural Principle License) for commercial and safety-critical deployment — full text in the attached PDF and at https://zenodo.org/records/20753406.

The Zenodo `license` field shows CC BY-NC 4.0 (the SPDX ID accepted by Zenodo); S-APL v2.0 is the second, mandatory license for any commercial or safety-critical use.

───────────────────────────────────────────────

Public reference specification defining the minimal requirements for any system to claim AMAGI-class compliance.

INTENDED AUDIENCE
Regulatory bodies, certification authorities, academic reviewers, and standards organizations evaluating hardware-enforced AI safety architectures. This document is the public-facing compliance gateway to the AMAGI Architectural Class.

SCOPE
Normative specification of the six non-droppable invariants (INV-001 through INV-006), the irreducible Trusted Computing Base (≤500 lines HDL), and the mandatory audit protocols that any AMAGI-compliant implementation must satisfy. Conformance to this specification is the prerequisite for S-APL v2.0 licensing and AGIAM certification.

COMPLIANCE MODEL
Tier 1–6 certification levels per S-APL v2.0 Section 6, with Tier 1 (Reference Implementation) representing full hardware-enforced compliance and Tier 6 (Derivative Architecture under DALA) representing licensed third-party implementations. All tiers require preservation of the six invariants.

ALIGNMENT
This document is normatively aligned with: Framework v2.0, NIPU v1.1, MEM v1.1, NIST AI RMF Profile v1.1, Master Parameter Sheet v2.0, S-APL v2.0, and AGIAM Charter v1.2.

Files

Amagi_Open_Specification_v1.1.pdf

Files (301.0 kB)

Name Size Download all
md5:9af258f509fbe7bcd535aa8cf5916a9e
301.0 kB Preview Download

Additional details