AISM is the AI Sovereign Maturity Model: a way to capture technology risks and score them now, so they can be assessed over time. That is the whole idea, and it is worth stating plainly, because until this page existed there was no single place on this site that said it. The model has been covered inside other articles, in the CLI 0.8.0 remediation article and in the AI SAFE2 v3.1 article, but it has never had its own article, and the address /aism/ has returned a 404. This page fixes that. It is the concept home for AISM.
The judgment at the center of AISM is simple. Maturity is something an organization earns over time, through evidence, and a single number that mixes the organization, the product, and the claim proves none of the three. AISM keeps them apart and scores each one on its own terms, so that a decision maker can see what the organization can do, what a technology contributes, and what the evidence actually proves, without any of the three borrowing credibility from the others.
The problem: sovereignty claims without a score
Every AI vendor claims its product makes you sovereign, secure, or mature. Every benchmark claims to prove it. The claims arrive faster than any organization can test them, and the testing that does happen is rarely recorded in a form that can be revisited. Six months later, nobody can say what was assessed, against which controls, with what evidence, or by whom. The organization has bought the product and kept the uncertainty.
That is the problem AISM was built to solve. It does not test products for you. It gives you a disciplined way to record what a technology does, where it works, and which evidence supports the claim, and to score your organization's maturity separately from any of it. The score is a record that can be revisited, challenged, and improved. Without that record, every new product resets the conversation to zero.
What AISM is: three instruments
AISM is not one score. It is three review instruments, defined in the October 2, 2026 AISM Scoring Update, which also adopted the model's name. Each instrument answers a different question.
| Instrument | Question it answers |
|---|---|
| AISM Sovereignty Score | How mature is the organization, across five pillars and six dimensions? |
| Technology Contribution Profile v1.0 | What does this artifact address, implement, enforce, validate, evidence, or challenge, for named controls within a stated boundary? |
| Evidence Assurance v1.0 | What evidence supports this specific claim, and how strong, independent, and limited is it? |
The Sovereignty Score is the organizational instrument. It covers five pillars and six dimensions, and it measures the organization, not the product. The Technology Contribution Profile is the artifact instrument. It records, for a specific technology at a specific version and deployment boundary, its control effects, coverage, evidence, boundaries, and residual risk. Evidence Assurance is the claim instrument. It attributes strength, origin, independence, and limitations to each claim, one claim at a time.
The three stay separate by design. A technology contributes evidence. An organization earns its maturity level. The release states the rule directly: do not label a product "AISM Level 4" or treat released code as independent validation. A product cannot be mature. Only an organization can.
How the score works, and what it does not do
An AISM score depends on scope, evidence, assessor judgment, and methodology. It is not an automatic certification. That sentence, from the CLI 1.0.0 release, is the most important guardrail in the model. A score produced without a stated scope is a guess. A score produced without evidence is an opinion. A score produced without a named assessor is unaccountable.
The model is built to make the gaps visible rather than to hide them. Where a review has not established a score or an assurance rating, the result is "not assessed," and missing evidence stays visible. The CLI 0.2.0 release put it concretely: all 30 maturity cells remain unscored until a human assesses the evidence. No cell is filled by default, by inference, or by the vendor's say-so. A human reads the evidence and decides.
That discipline is what makes the score useful over time. A score taken now, with its scope and evidence recorded, can be compared with a score taken later. The organization can see whether it actually improved, or whether it only bought something new. That is the meaning of the concept: capture the risk and score it now, so it can be assessed over time.
What AISM connects to
AISM does not sit alone. It is the assessment layer of AI SAFE2, which retains 161 core controls, and it connects to the CLI at the point where assessment becomes action. The CLI 0.8.0 article on this site covers the first connection: AISM findings turned into evidence-bound remediation plans with owners, dependencies, alternatives, exit criteria, history, and residual risk. The CLI 1.0.0 release adds the command that does it, safe2 aism plan, which turns confirmed AISM findings into a reviewable remediation plan without granting execution authority.
The October 2026 scoring update added the review workflow that feeds those plans. The Technology Card is the working artifact: version, scope, controls, mechanisms, evidence, failure behavior, retained authority, and adoption conditions, recorded for one candidate at a time. Nine source-attributed examples show how to turn research and accountability cases into scoped acceptance work, without adopting unsupported scores. The assessment guide walks through the three instruments, the card workflow, and a worked example.
The model also has boundaries that should not be crossed. It documents how to review control mechanisms and evidence. It does not install an enforcement system, independently validate example technologies, or certify an organization. Existing CLI grades and Challenge Lab levels stay separate from the assurance rubric. Completing a Technology Card does not authorize execution, deployment, external notification, or a conformance claim.
What to do with it
Start with one control gap and complete a Technology Card for the candidate. Identify the gap and the artifact's exact version and deployment boundary. Copy the card and record each control effect, supporting artifact, coverage, uncertainty, and failure mode. Review the bounded ContractWarden design example in the repository, then apply the adoption rule with a named decision owner and acceptance conditions.
Then do it again next quarter, with the same scope, and compare. The comparison is the point. A single score is a snapshot. A series of scores, each with its evidence attached, is a record of whether the organization is actually getting more sovereign, or only getting more tools.
That record is what AISM exists to produce. The model has been working inside other articles and releases for months. It finally has its own page.
Sources
- AISM Scoring Update, October 2, 2026: https://github.com/CyberStrategyInstitute/ai-safe2-framework/releases/tag/2026-10-02_AISM_Scoring_Update
- AISM model README: https://github.com/CyberStrategyInstitute/ai-safe2-framework/blob/446e29a51c3dff75d3b4def9ce9a49c6cb13f077/AISM/README.md
- AISM assessment guide: https://github.com/CyberStrategyInstitute/ai-safe2-framework/blob/446e29a51c3dff75d3b4def9ce9a49c6cb13f077/AISM/ASSESSMENT-GUIDE.md
- Technology Contribution Profile v1.0: https://github.com/CyberStrategyInstitute/ai-safe2-framework/blob/446e29a51c3dff75d3b4def9ce9a49c6cb13f077/docs/TECHNOLOGY-CONTRIBUTION-PROFILE.md
- Evidence Assurance v1.0: https://github.com/CyberStrategyInstitute/ai-safe2-framework/blob/446e29a51c3dff75d3b4def9ce9a49c6cb13f077/docs/EVIDENCE-ASSURANCE.md
- CLI 1.0.0 release body: https://github.com/CyberStrategyInstitute/ai-safe2-framework/releases/tag/2026-10-5_CLI_1.0.0
- CLI 0.2.0 release body: https://github.com/CyberStrategyInstitute/ai-safe2-framework/releases/tag/2026-09-09_safe2_CLI_0.2.0
- AI SAFE2 CLI 0.8.0 article (AISM remediation): https://cyberstrategyinstitute.com/ai-safe2-cli-080-aism-remediation/
- AI SAFE2 v3.1 article: https://cyberstrategyinstitute.com/ai-safe2-v31-protocol-governance-boundary/
- Release notes audit, November 17, 2026: ~/workspace/seo-review/release-notes-audit-2026-11-17.md