Reporting & framework relationships

Agent-level reporting

An ATF report should identify:

Agent, version, owner and purpose.
Current and target level.
Requirement results by element.
MUST gaps and SHOULD exceptions.
Evidence and testing depth.
Open incidents and findings.
Promotion-gate position.
Decision, conditions, approvers and review trigger.

Do not report a level without its scope and basis.

Portfolio reporting

Portfolio views can show:

Agents assessed versus registered.
Current and target level distribution.
Readiness by element.
Agents held below requested authority.
Common MUST gaps.
Weak evidence or overdue tests.
Incidents, demotions and promotion reviews.

A portfolio average must not conceal a high-authority agent with a failed containment requirement.

Safe conclusion language

Prefer:

The selected agent was assessed against ATF Specification 0.9.1 for a target of L2 Junior. The current record supports maintaining L1 pending completion of action-boundary testing and the Security Validation gate.

Avoid:

“ATF certified.”
“Zero trust compliant” without scope and evidence.
“Safe for autonomous operation.”
“All agentic risks eliminated.”

Relationship with ACRS

ACRS characterises capability risk: what the system can do and how much impact that capability creates. ATF governs how authority is earned and constrained.

A high ACRS position should increase assurance and testing depth and may justify a lower initial ATF level. It does not mechanically decide a requirement result or promotion.

Relationship with MAESTRO

MAESTRO identifies layer-specific agentic threats. ATF establishes zero-trust governance controls that help prevent, detect and contain them.

Use MAESTRO findings to challenge ATF requirements. For example, tool and orchestration threats should inform action-boundary, attribution, injection-defence and containment tests.

Relationship with GTSAF

GTSAF provides the wider assurance baseline across governance, data, models, security, people, suppliers and operations. ATF focuses on runtime trust for a named agent.

Crosswalks support traceability and evidence reuse; they do not create automatic equivalence.

Relationship with other frameworks

NIST AI RMF provides the broader risk-management outcomes.
ISO/IEC 42001 provides the organisational AI management system.
ISO/IEC 42005 examines impacts on people and society.
EU AI Act and NAGF determine applicable legal and regulatory obligations.

ATF does not replace any of these. The agent may need all of them.

Governance review questions

Is requested autonomy proportionate to capability risk?
Do threat findings change the target level or testing plan?
Is every cross-framework evidence reuse verified for this agent?
Are incident and demotion trends visible to accountable governance?
Does the report distinguish conformance assessment from certification?