Where AI governance meets operational reality.

FRAMEWORKS

Four frameworks. Different operational roles.

ODA3 Institute publishes distinct instruments for AI safety and security controls, incident description, incident response and Physical AI. They establish context for examination, not a finding about a particular deployment.

FRAMEWORK RELATIONSHIPS

How the frameworks work together.

These instruments connect control context, physical effects, incident records and response decisions. This is a conceptual information flow, not an operating software integration or a mandatory sequence for every use.

01

GAISSF™ · CONTROL CONTEXT

Identify the requirement, applicable control and evidence expectation.

02

PAI-SF™ · PHYSICAL AI

Where physical effects are possible, examine the relevant sensing, autonomy, command and action controls.

03

UAIF™ · INCIDENT RECORD

Structure a material AI incident with its identity, causal context, impact and supporting records.

04

AI-IRF™ · RESPONSE

Use the incident context for investigation, containment, recovery and operational learning.

Findings and lessons can inform review of the relevant GAISSF™ and PAI-SF™ controls. Preserve the source, verification state and limits of each record.

Explore platform interoperability →

THE ECOSYSTEM

Choose the instrument by the question.

The four frameworks are complementary, not interchangeable. Start with the decision or record you need to structure, then consult the applicable specification.

CONTROL CONTEXT

GAISSF™

Which safety and security controls are relevant?

Use the control framework to define the relevant safeguards and the evidence expected for examination.

9 domains · 59 controls

INCIDENT DESCRIPTION

UAIF™

How should an AI incident be described?

Use the taxonomy and schema to structure an incident record. Classification does not itself verify that a reported event occurred.

3 profiles · 2 extensions

INCIDENT RESPONSE

AI-IRF™

How should an AI incident be handled?

Use the lifecycle to organise preparation, investigation, containment, recovery and learning, with the applicable controls and evidence.

19 domains · 79 controls

PHYSICAL AI SECURITY

PAI-SF™

What changes when AI acts in the physical world?

Use the Physical AI security framework to examine controls where AI senses, decides and acts, including kinetic effects.

12 domains · 41 controls

Counts describe the published framework scope, not a blanket requirement for every system. Apply the version-specific scope and applicability rules in the controlling publications.

FROM FRAMEWORK TO EVIDENCE

A requirement does not prove implementation.

A control tells an organization what to establish. A system claim requires records tied to a defined version, environment and period. Provenance and verification determine whether those records can support a finding.

Define

Identify the requirement, system context and relevant control objective.

Substantiate

Examine implementation records and provenance; verify evidence against defined criteria.

Conclude

State the finding, evidence gaps and reliance limits for the particular system.

Explore the evidence discipline →

MAPPINGS & CROSSWALKS

Show relationships without claiming equivalence.

A mapping states a documented relationship under its scope. The result may be Supported, Partial, Gap or Unknown. Conformity requires the applicable obligation, actual implementation and supporting evidence.

Supported

Relationship supported for the stated scope.

Partial

Some but not all elements addressed.

Gap

No sufficient element under the defined mapping.

Unknown

Insufficient basis to determine the relationship.

For commissioned specifications, mappings or technical submission support:

Explore Standards Development services →

PUBLICATION CONTROL

Use the versioned specification.

Framework titles and counts are orientation. Before use, check the document ID, version, publication status, applicable scope and usage terms in the primary publication. Supporting crosswalks and guides may have a different status from the framework standard. The controlling records are maintained at docs.oda3.org.

CURRENT BOUNDARY

Published frameworks; distinct assurance work.

Publishing, using or mapping a framework does not establish regulatory compliance, effectiveness in deployment, certification or independent conformance. The evidence model is being tested and machine-readable structures remain in development.

UAIF™ V1.0

Three profiles. Two additive extensions.

Three profiles

  • Core
  • Enterprise
  • Regulatory

Two additive extensions

  • SOC_SIEM
  • AI_Security

Final Publication describes the specification. It does not establish regulator endorsement, production deployment maturity, sector calibration or independent inter-rater evidence.

Conformance test suite: developed; not yet published, as stated on the current framework page. Consult the documentation for the applicable release and readiness record.

Open the UAIF documentation ↗