ai-safe2-framework

Technology profile repository coverage review

Delivery record and implementation boundaries

AI SAFE² Review

Framework Home Cross-Pillar Governance AISM NEXUS Technology Profile

Documentation updated: 2026-10-02. TCP and Evidence Assurance v1.0 are CSI assessment methods for use and feedback. All AISM pages include a dated change summary and link the practical assessment guide. Release announcements are prepared outside the repository for owner publication.

Canonical-home update (2026-10-13)

The complete TCP, Evidence Assurance, Technology Card, validation, incident-review, and repository-coverage documents now live under AISM/. Current repository links and machine-discovery pointers use these paths. The technology evidence request supports provider and internal-owner intake before claim review. Earlier pinned revisions preserve the original paths; unpinned links to the former docs/ files must be updated. The record below describes the original introduction; this relocation changes no method version, rubric, organizational formula, or runtime contract.

Outcome and acceptance boundary

The current full name is AI Sovereign Maturity (AISM) Model, shortened to AISM or AISM Model. Current headings, root navigation, scoring/assessment prose, and naming guidance use this expansion. Historical submission wording and asset filenames remain intact with a current-name pointer; the Sovereignty Score and Sovereignty Matrix retain their distinct names. This is an editorial naming change with no scoring, schema, command, or path migration.

This method separates organizational AISM maturity, scoped artifact contribution, and claim-level assurance throughout repository guidance. Acceptance requires a central review method, reusable card, all nine supplied example dispositions, canonical control references, actionable validation/incident practices, root navigation, machine discovery, and clear compatibility/implementation limits.

Base: e66182e93bef1debe61c2cc17f089c60d16829a9, fetched from the repository’s default main branch on October 2, 2026. Work is isolated on docs/technology-contribution-profile. The PR head commit is the exact revision under review; later changes require affected checks to be rerun.

Before: organizational methodology, CLI numeric E grades, and Challenge Lab C levels existed, but no dedicated artifact contribution/card/assurance-v1.0 review entry point existed. After: the linked method names each instrument and preserves their distinct meanings. This is a document/path comparison, not an experiment demonstrating improved deployed security.

Coverage review

Repository surface Changes or reviewed disposition
Root README, quickstart New three-instrument summary, method/card/examples links, score interpretation
Agent entry point, manifest Discovery pointers, method/implementation status, explicit separation from organizational maturity and CLI grades
Shield Actual interception/judgment/actuation boundary, composed paths, side effects
Ledger Evidence origin, protected records, reconstruction, retention and notification ownership
Circuit Breaker Actuation/stopping authority, direct/descendant bypass, outage and revocation tests
Command Center Probabilistic judgments, forged process evidence, accountable notification
Learning Engine Paired testing, benchmark limits, learned-policy integrity and owned regressions
Cross-Pillar Governance Applicable CP.5 profile/plane, CP.6 feedback, CP.10 rights/stopping authority
AISM README, methodology, assessment, maturity ladder Organizational subject retained; scoped technology evidence cannot assign organizational cells or levels automatically
Crosswalk, threat matrix, control stack, operational loop Procurement rule, evidence effect vocabulary, boundary tests, owned incident/policy-learning intake
AISM strategic architecture, sovereignty matrix Reviewed organizational governance/control/autonomy context; no axis, architecture, or formula change needed
NEXUS, Gateway Same claim discipline for first-party implementations; actual tested version and plane; native risk routing separate
Scanner, Skills, MCP server Static/inspection evidence separate from runtime enforcement; native verdict/score semantics retained
CLI, scope guide, readiness No TCP command or payload; existing intake/unscored cells, E weights/caps/categories, and human release authority retained
Dashboard Proposed future presentation boundary; no new TCP view or relabeling of current scores
Examples index, research index Method/card pointers, nine-example review; no runnable integration added or generated table hand-edited
Challenge Lab, harness design, CLI guide C scale distinct, new test candidates explicit; frozen Challenge 001 and current executable scope untouched
Contribution guidance, PR template Claim/control/boundary evidence requirements and separate proposal/result states
Evolution, CLI roadmap, brief triage, UX standard Assessment methods recorded; runtime work/version migration deferred; navigation and terminology rules linked
Core/profile datasets, NEXUS schemas, scanner rules, CLI schemas/scoring, dashboard generated data Reviewed relevant IDs, versions, counts, discovery, grade/cap semantics; no contract change required for Markdown review instruments
Historical notes, release notes, published assets Kept as original records; new source-reviewed guidance is discoverable from current indexes rather than rewriting historical evidence

Delivered documents and deferred work

The actual kernel/runtime integrations, new benchmark experiments, policy-memory governor, production evidence collection, TCP machine schema/CLI/dashboard, and grade translation are deferred implementation proposals with acceptance evidence listed in the backlog. No live offensive evaluation or external notification was performed. No organization or candidate was certified or assigned a maturity level.

Risks, compatibility, and ownership

The principal risk is readers converting a capability, source claim, or E rating into maturity or certification. The documents counter that through explicit subject, effect, boundary, assurance rubric/version, provenance, independent review status, and unassessed states. Producer claims remain attributed; the notification count remains unverified. Examples’ supplied numeric ratings are not promoted without their underlying evidence.

Framework controls remain 161; CP.5.MCP remains 19 profile controls; UAS remains 27 profile requirements. Existing versions, formulas, numeric E weights, verification caps, schemas, and frozen challenge artifacts remain unchanged. Manifest additions are discovery-only, not schema catalog or executable commands.

CSI maintains the v1.0 assessment methods and welcomes evidence-based feedback; deployment/incident owners make adoption, risk, retention, and notification decisions. No decision authority is delegated by this change. Rollback is a revert of the documentation/discovery commit; no runtime state or data migration exists. Independent review and hosted CI status belong to the PR, not to a fabricated assurance score for this self-review.

Verification

Run from the repository root:

python scripts/check_repo_ux.py
python scripts/generate_examples_table.py --check
python scripts/check_agent_manifest.py
python -m pytest tests/ scanner/tests/ -q

Check added local Markdown destinations/anchors and canonical control IDs as part of document review. Hosted Linux and supported Python-matrix checks remain separate from local Windows results. The PR records actual outcomes and any baseline or environment limitations at the tested revision.


Technology Profile Evidence Assurance Examples

AI SAFE² v3.1 · Cyber Strategy Institute