How HI-AAF is developed.

HI-AAF is developed by Human Intelligence as an independent framework. We make our editorial process public so that readers can judge the rigor of the document and so that the conversation about agent assurance can be had in the open.

§ 1 · Editorial process

HI-AAF is authored by a small editorial team at Human Intelligence with input from external reviewers drawn from practising AI risk leads at customer organizations, academic researchers, and standards-body participants. The editor in chief approves the published edition.

Each material change to the framework follows a documented cycle: a proposal is drafted with a written rationale and dissent log, circulated to the external review panel for comment, revised, and then either incorporated into the next edition or formally rejected with reasons. Rejected proposals are preserved.

§ 2 · Review cadence

Minor editions (e.g., v1.1 → v1.2) are released no more often than quarterly. Major editions (e.g., v1 → v2) follow a deeper review and are released no more often than annually. Out-of-cycle releases are reserved for vulnerabilities in the framework itself — for example, a control whose application has been demonstrated to be unsafe in production.

Customers under active Maintenance are notified of material edits with a documented transition window before re-attestation is required.

§ 3 · Proposing a change

Proposals for changes to HI-AAF — new controls, edits to existing controls, additions to the glossary, or corrections — are accepted by written submission to standard@humanintel.net. Submissions should state the control or term affected, the proposed change, the rationale, and (where possible) evidence from production practice.

Anonymous submissions are accepted; submissions from named authors may be cited in the resulting edition's acknowledgments. Submissions are reviewed at the next editorial cycle and the author is notified of the outcome.

§ 4 · Conflicts of interest

Human Intelligence is the publisher of HI-AAF and also the firm that performs Assessments against it. This is a structural conflict that the editorial process is designed to mitigate. Specifically: editorial decisions are not made by individuals with active commercial engagements with affected customers; external reviewers are drawn from outside HI; and Assessment outcomes are not contingent on the customer's commercial relationship with HI.

We acknowledge that this mitigation is not equivalent to the separation a regulator enjoys. We believe the public editorial process is the appropriate response at this stage; we will revisit the structure if and when the framework matures into a position where independent governance becomes practicable.