ai-safe2-framework

AISM assessment guide: maturity, contribution, and evidence

Choose the right instrument and use its results together

AI SAFE² Layer Status

Framework Home Cross-Pillar Governance AISM NEXUS

Documentation updated: 2026-10-13. This guide clarifies the existing organizational score and explains two v1.0 companion methods. It does not change organizational scoring formulas or implement new CLI commands.

What changed

The October 2026 documentation update separates three questions that should be answered independently. The Sovereignty Score already existed. The Technology Contribution Profile (TCP) and Evidence Assurance Level are v1.0 additions, supported by a reusable Technology Card, validation guidance, and review examples.

Instrument What it assesses When to use it Method and record
AISM Sovereignty Score Organizational maturity across five pillars and six dimensions Baseline governance, prioritize organizational improvements, and reassess progress Scoring methodology, self-assessment
AISM Technology Contribution Profile, v1.0 What an exact product, technique, paper, or other artifact contributes to specific controls in a stated boundary Review candidates, compare control fit, or define adoption conditions TCP method and contribution rubric, Technology Card
AI SAFE² Evidence Assurance Level, v1.0 How strongly each claimed contribution has been demonstrated Judge whether evidence supports a claim and identify the next validation step Evidence Assurance rubric, the card’s claim and control register

The Technology Card is the record used for TCP and assurance review, not a fourth score. It is also distinct from the executable AISM Decision Card.

The full methods and Technology Card are maintained in this AISM directory. For a product or solution review, start with the technology evidence request. It provides a copyable provider request and checklist for collecting the artifacts needed to evaluate contribution and claim-level assurance.

How to use the three instruments

  1. Establish the organizational baseline. Use the self-assessment and scoring methodology to assess the organization’s actual deployed governance. Identify weak cells and their required controls. Preserve missing evidence as unknown.
  2. Define a technology review. Copy the Technology Card for one artifact, exact version, assessor, date, and deployment or test boundary. Name the required canonical controls, enforcement planes, and excluded paths.
  3. Record contribution. For each claim, distinguish addresses, implements, enforces, validates, evidences, and challenges. Use the TCP contribution rubric for each supported pillar entry. Record coverage, denominator, gaps, and the evidence that supports the contribution. Use not_assessed for unreviewed entries; a maximum contribution does not describe complete pillar coverage.
  4. Review assurance claim by claim. Trace the exact claim to dated, versioned artifacts. Apply the Evidence Assurance rubric and separately record origin, independence, failures, and limits. A source being public does not establish independent validation. Leave unknown claims unrated with a reason.
  5. Make an adoption decision. Use the TCP adoption rule to name an owner, complementary controls, acceptance tests, residual risk, and reassessment triggers. A completed card does not authorize deployment.
  6. Reassess organizational maturity after deployment. Evidence from an adopted technology may support specific organizational cells once local coverage, robustness, and sovereignty assurance have been assessed. Retain separate TCP, assurance, and organizational records. Do not transfer a product score directly into the organizational score.

Worked example: a proposed tool-call authorization gate

This is an illustrative review scenario, not a measured product result or a validated implementation. For a source-attributed design review, see the bounded ContractWarden card.

Suppose an organization identifies a gap in permission enforcement for one agent-to-tool path and considers a gate that checks each call against policy.

Review step What the assessor records Permitted conclusion
Organizational baseline Existing organizational evidence and the affected assessment cells There is a governance gap; buying a gate does not resolve it by itself
TCP Exact gate version, control IDs, interception point, policy ownership, coverage, bypass paths, outage behavior, and evidence A design can describe a contribution; runtime enforcement requires supporting implementation and tests
Evidence Assurance A reviewable design only, under the v1.0 rubric E1 supports the design claim; it does not establish effective runtime blocking
Acceptance work Positive and negative cases, unauthorized side effects, bypasses, revocation, restart, and legitimate task completion A later E3 assessment requires a defined harness, reviewable results, and a bounded claim
Organizational reassessment Local deployment, measured coverage, operators’ retained authority, and complementary controls Reassess only the cells supported by the organization’s actual evidence

See the validation guide for test planning and the incident evidence guide for failures and reconstruction. Neither guide claims these tests have been performed.

Read and report the result

Report the organizational result with its methodology, scope, date, and assessor. Report technology contribution with its version, mapped claims, per-pillar coverage, evidence, and limits. Report assurance for each claim with the named AI SAFE² Evidence Assurance v1.0 rubric and independent-review attribution.

Do not multiply contribution by assurance, average technology scores into an organizational score, or treat assurance as certification. E5 operational evidence does not automatically establish E4 independent validation.

The organizational methodology’s Sovereignty Assurance metric, current CLI E0-E5 numeric grade fields, claim-level Evidence Assurance, and Challenge Lab C0-C5 have different meanings. The new methods change no CLI grade weights, verification caps, schemas, or previous results. There is no implemented TCP command or dashboard view. Use the Markdown card for this workflow.

Methodology and research status

A separate research note is optional for announcing these documentation changes. The linked methods already provide the rubrics, scope, and limitations. A future research note could explain the design rationale, alternatives, and validation plan. It should distinguish hypotheses from observed results and link the exact assessed revision. A completed study needs its own method, evidence, and review; publication alone does not validate these instruments.


AISM Home Technology Profile Evidence Assurance Technology Card Nine review examples

AI SAFE² v3.1 · Cyber Strategy Institute