AI Management Framework

Published together with its own measurement record — the controls, the anchors, the scoring protocol, and the measured evidence of how far two independent assessors actually agree, including where they do not.

A control-based instrument for assessing and improving how an organisation governs, manages, maps and measures its AI systems. Version 1.0.12 · published by PraeSyn, LLC · CC BY-NC-ND 4.0

Download the framework

The complete framework is one archive: 84 files of markdown source, the control catalogue, the question bank, the profiles, the mappings, the validation record and the licence. No form, no account, no email required.

AIMF-Full-Documentation-v1.0.12.zip

  • sha256 e99c07d1402f812c4618119bbab9cf16c6a583d46b2693af8c92aba9507c7e6d
  • Size 1,196,412 bytes
  • Version 1.0.12, released 21 October 2026
  • Also available as a tagged release with the same bytes attached

Verify the download with shasum -a 256 AIMF-Full-Documentation-v1.0.12.zip. If the hash does not match the value above, discard the file.

What the framework contains

  • 72 controls across four functions GOVERN, MANAGE, MAP and MEASURE, each with a definition, evidence expectation and mapping to recognised standards.
  • A five-level maturity scale ML1 to ML5 with per-control behavioural anchors, so two assessors given the same evidence can reach the same score.
  • A question bank one question, one evidence expectation and five anchors for every control.
  • Six selection profiles Foundational, Balanced, Healthcare, Financial Services, Government and High Assurance, plus five maturity-level guides.
  • Standards mappings NIST AI RMF 1.0, ISO/IEC 42001:2023, ISO/IEC 27001:2022 (partial — see below) and Executive Order 14179.
  • The measurement record the instrument-validation study, the held-out case, the scoring protocol and the shipped checking tool.

What this framework is not

These are measured results and stated scope limits, not caveats.

  • The ISO/IEC 27001:2022 mapping is Partial. 54 of the 72 controls carry an ISO 27001 mapping tag; 18 are recorded as no-analogue. The mapping is not clause-complete and is not a statement of conformity with ISO/IEC 27001:2022.
  • Reproducibility holds within one model family, not across vendors. On the held-out case AIMF-VAL-2026-B, two independent blind assessments by two sessions of one model agreed on 65 of 72 controls (90.3%). Two different vendors agreed on 43 of 72 (59.7%). Quote 65 of 72 with that qualifier.
  • AIMF is not a certification. It is an assessment instrument. No certificate, audit opinion or ISO certification is issued or implied, and neither PraeSyn nor the framework is an accredited certification body.

How PraeSyn delivers it

The Balanced AIMF Assessment assesses one organisation against one profile, using this framework and its published scoring protocol, and produces a scored control-by-control record, a board-readable gap report, a customer-owned implementation profile and a 90-minute readout — fixed scope, fixed fee, two weeks elapsed.

Scope, the number of controls assessed and the fee are fixed in a written proposal before work starts.

Talk to us about an assessment

Licence and attribution

The AIMF is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International licence (CC BY-NC-ND 4.0). Commercial consulting and implementation services using the framework are permitted with attribution.

Attribution required

This work uses the AI Management Framework (AIMF) by PraeSyn, LLC,
licensed under CC BY-NC-ND 4.0.
https://www.praesyn.com/aimf

For academic or professional citations

PraeSyn, LLC. (2026). AI Management Framework (AIMF), Version 1.0.
https://www.praesyn.com/aimf. Licensed under CC BY-NC-ND 4.0.

The full licence text, including the proprietary-component carve-outs and the commercial-use clarifications, ships inside the bundle: LICENSE.md. The licence deed itself is at creativecommons.org.

Commercial licensing and proprietary-tooling enquiries: licensing@praesyn.com.

Source and provenance

The framework's source is public. This release was gated: the tag is cut on the commit that carries its own publication gate record, and the archive above is byte-identical to that commit's documentation tree.