ATF Agent Maturity Model

March 13, 2026 · View on GitHub

The Agentic Trust Framework uses a progressive maturity model that matches agent autonomy to demonstrated trustworthiness. Agents don't automatically receive higher autonomy levels; they earn promotion through performance, security validation, and business value delivery.

The metrics and timeframes below are illustrative. Organizations should calibrate thresholds based on regulatory context, risk tolerance, and domain criticality.

This model aligns with the AWS Agentic AI Security Scoping Matrix, providing enterprise security alignment.

The Four Levels

LevelNameAutonomyHuman InvolvementAWS Scope Alignment
1InternObserve + ReportContinuous oversightScope 1 (No Agency)
2JuniorRecommend + ApproveHuman approves all actionsScope 2 (Prescribed Agency)
3SeniorAct + NotifyPost-action notificationScope 3 (Supervised Agency)
4PrincipalAutonomousStrategic oversight onlyScope 4 (Full Agency)

Level 1: Intern Agent

Definition: Intern agents operate in read-only mode. They can access data, perform analysis, and generate insights, but cannot take any action that modifies external systems.

Capabilities

AllowedNot Allowed
Read data from authorized sourcesCreate, update, or delete records
Analyze and process informationSend communications
Generate reports and summariesTrigger workflows or automation
Flag items for human attentionAccess credentials or secrets
Answer questions about dataModify any external system

Governance Requirements

Core ElementRequirements
IdentityBasic authentication, role assignment, audit logging
BehaviorComprehensive action logging, output review
Data GovernanceInput validation, PII detection, output filtering
SegmentationRead-only resource access, strict allowlists
Incident ResponseCircuit breaker, kill switch, human escalation

Use Cases

  • Security log monitoring and alert triage
  • Customer sentiment analysis
  • Document summarization and search
  • Data quality assessment
  • Compliance monitoring and reporting

Risk Profile

Lowest risk. Intern agents cannot cause direct harm through action. Primary risks:

  • Information disclosure (if outputs shared inappropriately)
  • Incorrect analysis leading to poor human decisions
  • Resource consumption (compute, API calls)

Minimum Time at Level

2 weeks of operation before promotion eligibility.


Level 2: Junior Agent

Definition: Junior agents can recommend specific actions with supporting reasoning, but require explicit human approval before any action is executed.

Capabilities

AllowedNot Allowed
All Intern capabilitiesExecute actions autonomously
Generate action recommendationsApprove other agents
Provide reasoning for recommendationsModify security settings
Draft content for human reviewAccess resources outside scope
Prepare transactions for approval
Execute actions after human approval

Governance Requirements

Core ElementRequirements
IdentityOAuth2/OIDC for approval workflows, session context
BehaviorBehavioral baseline, anomaly flagging, reasoning capture
Data GovernancePrompt injection detection, output validation, data lineage
SegmentationAction allowlists, transaction limits, rate limiting
Incident ResponseAutomated alerting, basic rollback, approval queue pause

Use Cases

  • Customer service response drafting
  • Purchase order preparation
  • Meeting scheduling assistance
  • Code review and suggestions
  • Marketing content creation

Risk Profile

Low risk. Human approval gates all impactful actions. Primary risks:

  • Approval fatigue leading to rubber-stamping
  • Incorrect recommendations that humans don't catch
  • Queue backlogs during high volume

Minimum Time at Level

4 weeks of operation with >95% recommendation acceptance rate before promotion eligibility.


Level 3: Senior Agent

Definition: Senior agents can execute actions within defined guardrails and notify humans of what they did and why. They operate with significant autonomy but maintain transparency through real-time notifications.

Capabilities

AllowedNot Allowed
All Junior capabilitiesModify own permissions
Execute approved action types autonomouslyOverride security controls
Send notifications to stakeholdersEscalate other agents
Trigger downstream workflowsOperate outside defined scope
Access credentials within scope
Coordinate with other agents (within limits)

Governance Requirements

Core ElementRequirements
IdentityAttribute-based access, just-in-time privileges, mutual TLS
BehaviorReal-time anomaly detection, sequence analysis, intent drift detection
Data GovernanceFull data classification, source verification, lineage tracking, output governance
SegmentationPolicy-as-code enforcement, temporal boundaries, cascade prevention
Incident ResponseIsolation capability, checkpoint/resume, graceful degradation

Use Cases

  • Infrastructure auto-scaling
  • Automated customer refund processing (within limits)
  • Routine IT ticket resolution
  • Inventory reordering
  • Scheduled report distribution

Risk Profile

Moderate risk. Autonomous execution creates exposure. Mitigations:

  • Real-time notifications enable rapid human intervention
  • Transaction limits cap individual action impact
  • Cumulative limits prevent slow-rolling attacks
  • Graceful degradation to Junior level on issues

Minimum Time at Level

8 weeks of operation with zero critical incidents before promotion eligibility.


Level 4: Principal Agent

Definition: Principal agents operate autonomously within an approved domain, escalating edge cases rather than routine decisions. They represent the highest capability and highest governance burden.

Capabilities

AllowedNot Allowed
All Senior capabilitiesModify governance policies
Self-directed execution within domainPromote other agents
Dynamic boundary negotiation (within policy)Operate outside defined domain
Escalate edge cases to humansBypass security controls
Coordinate complex multi-agent workflows
Request temporary privilege elevation

Governance Requirements

Core ElementRequirements
IdentityHardware-bound identity, full policy-as-code, privilege attestation
BehaviorContinuous scoring, real-time explanation, autonomous escalation
Data GovernanceSource trustworthiness scoring, full lineage graphs, quality gates, regulatory compliance
SegmentationFull microsegmentation, real-time policy evaluation, dynamic boundaries
Incident ResponseAutomated detection/containment/recovery, novel incident escalation

Use Cases

  • Algorithmic trading within risk parameters
  • Autonomous security incident response
  • Complex supply chain optimization
  • Self-healing infrastructure management
  • Multi-system business process automation

Risk Profile

Highest governance requirements. Full autonomy demands maximum controls:

  • Continuous behavioral monitoring for drift
  • Real-time anomaly scoring with automatic containment
  • Complete audit trail for regulatory review
  • Regular security validation and penetration testing

Minimum Time at Level

Ongoing. Principal agents require continuous validation. Any significant incident triggers automatic demotion.


Promotion Criteria: The Five Gates

For an agent to be promoted to the next level, it must pass all five gates:

Gate 1: Performance

Demonstrated accuracy and reliability over the evaluation period.

MetricIntern → JuniorJunior → SeniorSenior → Principal
Minimum Time at Level2 weeks4 weeks8 weeks
Accuracy RateN/A>95%>99%
Availability>99%>99.5%>99.9%
Response Time SLAMetMetMet

Gate 2: Security Validation

Passes security audit appropriate to target level.

RequirementJuniorSeniorPrincipal
Vulnerability Assessment
Penetration Testing
Adversarial Testing
Code Review
Configuration Audit

Gate 3: Business Value

Measurable positive impact demonstrated.

RequirementJuniorSeniorPrincipal
Defined Success Metrics
Baseline Established
Improvement Demonstrated
ROI Calculation
Stakeholder Sign-off

Gate 4: Incident Record

Clean operational history at current level.

RequirementJuniorSeniorPrincipal
Zero Critical Incidents
Minor Incidents Resolved
Root Cause Analysis CompleteIf applicable
Remediation VerifiedIf applicable

Gate 5: Governance Sign-off

Explicit approval from authorized stakeholders.

RequirementJuniorSeniorPrincipal
Technical Owner Approval
Security Team Approval
Business Owner Approval
Risk Committee Approval
Documentation Updated

Demotion Criteria

Agents can be demoted at any time if they fail to maintain standards.

Automatic Demotion Triggers

TriggerResult
Critical incident at current levelImmediate demotion to Intern
Security vulnerability discoveredDemotion pending remediation
Repeated minor incidents (3+ in evaluation period)One level demotion

Review-Based Demotion

  • Performance metrics fall below threshold
  • Business value no longer demonstrated
  • Scope or purpose changes significantly
  • Underlying model or system changes

Demotion Process

  1. Incident/trigger documented
  2. Agent isolated (if immediate risk)
  3. Root cause analysis conducted
  4. New level determined
  5. Governance controls updated
  6. Agent reactivated at new level
  7. Re-promotion requires full gate passage

Operating Model

Roles and Responsibilities

RoleResponsibility
Agent OwnerDay-to-day operation, performance monitoring, issue resolution
Technical OwnerArchitecture, implementation, technical governance
Security TeamSecurity validation, penetration testing, incident response
Business OwnerUse case definition, value measurement, stakeholder management
Governance BoardPromotion/demotion approval, policy setting, risk acceptance

Review Cadence

ActivityFrequency
Performance ReviewWeekly
Security ScanWeekly
Promotion Eligibility CheckMonthly
Full Security AuditQuarterly
Governance ReviewQuarterly
Penetration TestAnnually (Senior+)

Quick Reference: Deployment Checklist

Before Deploying Any Agent

  • Agent registered in identity system
  • Unique identifier assigned
  • Ownership documented
  • Purpose and scope defined
  • Initial maturity level set (always Intern)
  • Governance controls configured
  • Monitoring enabled
  • Incident response procedures defined
  • Kill switch tested

Before Promoting to Junior

  • 2+ weeks at Intern level
  • All outputs reviewed, accuracy confirmed
  • Approval workflow configured
  • Approver roles assigned
  • Security scan passed
  • Business owner sign-off obtained

Before Promoting to Senior

  • 4+ weeks at Junior level
  • >95% recommendation acceptance
  • Behavioral baseline established
  • Anomaly detection configured
  • Notification channels configured
  • Transaction limits set
  • Penetration test passed
  • Security team sign-off obtained

Before Promoting to Principal

  • 8+ weeks at Senior level
  • Zero critical incidents
  • >99% action accuracy
  • ROI demonstrated
  • Adversarial testing passed
  • Full audit completed
  • Risk committee approval
  • Executive sponsor sign-off


The Agentic Trust Framework is an open specification licensed under CC BY 4.0.