Skip to content

Reference & glossary

LabelMeaning
AgentA system component that pursues goals and can use delegated capability to observe, recommend or act
ElementOne of Identity, Behavior, Data Governance, Segmentation or Incident Response
RequirementOne of the 25 canonical ATF control statements
MUSTRequired for conformance at the selected maturity level
SHOULDExpected unless a justified and reviewed exception exists
MAYOptional in ATF at that level; other obligations can still require it
Current levelPresently approved authority
Target levelProposed future authority; not an approval
Promotion gateEvidence checkpoint for performance, security, value, incidents or governance
HoldMaintain present authority but do not promote
DemotionReduce authority because current trust conditions are not met
Privacy ModeReduced AI-assistance context without free text or direct identifiers

No. It is an open agent-governance specification.

The canonical repository currently describes Specification 0.9.1 as a Public Review Draft. Verify the repository before making a time-sensitive version claim.

Does completing all 25 items make an agent safe?

Section titled “Does completing all 25 items make an agent safe?”

No. The assessment supports a bounded governance decision. Threat modelling, impact assessment, legal review, operational monitoring and incident readiness remain necessary.

No. It means greater approved autonomy and therefore a higher control burden. Many agents should remain at Intern or Junior.

Sequential promotion is the defensible default because each level supplies operating evidence. Avoid skipping the observation and review periods.

No. A target-level MUST must be Fully met. Partially met and Not addressed are blockers.

No. It provides saved, per-requirement advice for human review.

This assessment used ATF Specification 0.9.1 and Conformance 0.9.0, identified by the canonical project as a Public Review Draft. Results are agent-specific and do not represent certification.