Updates from
the lab.
A running record of framework releases, trademark filings, publications submitted, and the institutional steps that build out the certification authority. Newest first.
White paper revised to v3.7
The white paper on SSRN is replaced with a v3.7 revision, 68 pages, bringing the preprint level with the current standard. The version it replaced described the framework at v3.0. The revision carries the failure-defined principle, the two-tier requirement structure, and the co-located verification approach as they now stand. SSRN retains no prior version, so the superseded paper is gone.
Read the revised white paper on SSRN →Framework v3.7 released
The current published revision, now the standard SCL assesses against, under a new citable DOI. A new Section 1.6 Conformity Basis states what the certification determination rests on, and defines the certified configuration: boundary and classification, the operational design domain, the declared thresholds, the declared attack pattern sets, and the security baseline. The normative reference list is cut from ten documents to three, with the remainder moved to an informative table, so no external standard is levied unless a requirement statement names it and states the extent. A new requirement, AI-7.7 Artifact Load Safety, covers every path by which an artifact is admitted through a loader. Appendix D.5 is settled to two outcomes, Certified or Not Certified, with no intermediate result. The thirteen requirement areas (AI-1 through AI-13) and the three tier classification structure carry forward from v3.6.
Download v3.7 (DOI) →Paper accepted to AIAA SciTech Forum 2027
The abstract submitted in May was accepted on August 24, 2026 into session IS-05 of the Intelligent Systems: Space Trusted Autonomy track. The final manuscript is due December 1, 2026, with presentation scheduled for January 11, 2027 at the forum in Orlando, Florida.
Framework v3.6 released
Point revision published under a new citable DOI. This release makes the applicability determination verifiable. Whether the framework applies now turns on a single anchor criterion, whether a component's correctness can be fully verified against explicitly written specification by established methods, with the list of mechanisms treated as illustrative rather than load bearing. Each determination declares its boundary of analysis, and a normative deletion test for data fitted parameters separates learned components from conventional engineering methods that only resemble them. The thirteen requirement areas (AI-1 through AI-13) and the three tier classification structure carry forward from v3.5.
Download v3.6 (DOI) →Framework v3.5 released
Point revision published under a new citable DOI. The thirteen requirement areas (AI-1 through AI-13) and the three tier classification structure carry forward unchanged.
Download v3.5 (DOI) →Framework v3.4 released
Point revision published under a new citable DOI. The thirteen requirement areas (AI-1 through AI-13) and the three tier classification structure carry forward unchanged.
Download v3.4 (DOI) →Open verification demonstration on NASA turbofan data
A public, reproducible demonstration applies the framework's verification methods to NASA's C-MAPSS turbofan degradation dataset. A series of experiments exercises data partitioning, drift monitoring, out-of-distribution detection, and calibration, with the full pipeline, measured results, and figures published as an open repository under an MIT license. It accompanies a methodology manuscript now under review at a Nature Portfolio journal.
View the demonstration repository →Framework v3.3 released
Point revision published under a new citable DOI. The thirteen requirement areas (AI-1 through AI-13) and the three tier classification structure carry forward unchanged.
Download v3.3 (DOI) →Framework v3.1 released: refinements and clarifications
Point revision published under a new citable DOI. The thirteen requirement areas (AI-1 through AI-13) and the three tier classification structure carry forward unchanged from v3.0, with refinements to requirement wording, verification guidance, and cross references throughout.
Download v3.1 (DOI) →White paper posted on SSRN
The AI Requirements Framework white paper is publicly posted on SSRN as a citable preprint. It presents the failure-defined principle, the two-tier requirement structure, and the co-located verification approach in a single reference for practitioners and reviewers.
Read the white paper on SSRN →Website refresh and source document library launched
safetycriticallabs.com refreshed across every page in the cream and blue palette. A new Documents library publishes the full set of standards, regulations, and reference documents that inform the framework, with direct links to each official publisher.
View the document library →Framework v3.0 released: Architecture and Paradigm Requirements
Major structural revision. New Section 3 adds three architecture and paradigm requirement sets: AI-11 Multi-Model Systems, AI-12 Neural Networks, and AI-13 Continuous Learning and Adaptation. These apply conditionally based on system design and resolve prior lexicon contradictions by separating normative architecture requirements from informative implementation patterns.
Download v3.0 (DOI) →Framework v2.1.3: Deployment Format Validation
Added AI-4.7 Deployment Format Validation covering model transformations (quantization, pruning, knowledge distillation, framework conversion). Added Section 4.4 Continuous and Periodic Verification and a Cadence column to the Verification Matrix. Editorial pass on Sections 1 through 3 for terminology and lexicon consistency.
Abstract submitted to AIAA SciTech Forum 2027
Framework abstract submitted to the Intelligent Systems: Space Trusted Autonomy track at the AIAA SciTech Forum 2027. Decision notification expected on or about August 24, 2026. First of a multi-venue formal publication push that also targets JAIS, SAFECOMP, and a NIST companion piece.
ANAB accreditation conversation opened
Intake response submitted to the ANSI National Accreditation Board, opening the formal conversation toward eventual ISO/IEC 17065 accreditation for certification bodies. ANAB accreditation is the path that transforms the SCL mark from a credible third-party determination into a regulatory instrument.
Framework v2.1.1: Human factors expansion
AI-9 Human and AI Teaming expanded with three new sub-requirements: AI-9.7 Operator Qualification, AI-9.8 Workload Management, and AI-9.9 Training Program Requirements. New AI-8.5 Public Disclosure Support covering external model cards, EU AI Act Article 13 user information, EO 14110 and OMB M-24-10 federal AI inventory, and EU AI Act Article 71 public database registration.
USPTO trademark applications: the mark on record
Applications filed with the United States Patent and Trademark Office under Class 42 (scientific and technological services), covering the Safety Critical Labs name and the SCL design mark for the logo and seal. Current status is not tracked on this site, and none of the marks is registered.
Logo and seal design finalized
Final mark design completed. The SCL seal is the visible identifier on certificates, the digital badge, and the verification page. Prerequisite for the design mark trademark filing that followed in the same month.
First Zenodo deposit, citable DOI registered
The AI Requirements Framework was deposited to Zenodo on 14 March 2026 and assigned a citable DOI. Cite the framework at the concept DOI 10.5281/zenodo.19024420, which always resolves to the most recent version. Thirteen versions have been deposited to date, from 14 March 2026 to 15 September 2026. Entries dated before 14 March 2026 carry document revision dates from the framework's own revision history; entries dated on or after it carry Zenodo deposit dates.
Framework v2.1, ODD and Privacy
Added AI-1.0 Operational Design Domain with Operational Claim and eight attribute categories, formalizing the certification scope envelope. Added AI-10 Privacy and Data Protection with seven sub-requirements. Section 1.3 introduced the Domain Standards Citation Convention. Mission Support tier renamed to Operational Support throughout. Document revision date. Version 2.1 was not deposited to Zenodo under that number; the corresponding deposits are v2.1.1 and v2.1.2, both dated 16 April 2026.
Framework v2.0, Risk Score Methodology
Added AI-4.6 Operational Validation Without Ground Truth and AI-9.6 Operational Role Verification. Introduced Appendix D Risk Score Methodology, the structured and reproducible scoring system that issues a quantified risk score alongside every certification determination. Document revision date. This revision was deposited to Zenodo on 10 April 2026.
Framework v1.0, initial development
Initial framework development. First draft of the SCL AI Requirements Framework, establishing the requirement areas that address AI-specific failure modes not covered by traditional software assurance practices. This is a development milestone from the document revision history, not a publication date. The framework was first deposited to Zenodo, and first received a citable DOI, in March 2026.