Skip to main content

Adapt (Phase 5)

The Adapt phase enables trust evolution over time. Review policy suggestions, handle approvals, and monitor trust recovery.

Access via Agent Detail → Adapt tab.

Sub-tabs

Approvals

Agent-specific pending approval requests:

FieldDescription
Request IDUnique identifier
OperationWhat needs approval
SessionOriginating session
RequestedTimestamp
TimeoutTime remaining
Trust ImpactScore change if approved/rejected

Actions:

  • Approve - Allow operation to proceed
  • Reject - Deny operation
  • View Context - See full governance decision details

For the organization-wide approval queue, see Approvals.

Insights

AI-generated insights for improving governance:

Policy Suggestions

Based on observed patterns, OpenBox suggests new policies or rules:

Suggested Rule: "PII Access + External API"
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Trigger: DATABASE_READ on PII tables
Condition: EXTERNAL_API_CALL within 60s (without approval)
Action: REQUIRE_APPROVAL

Confidence: 87%
Based on: 12 similar violations in past 30 days

For each suggestion:

  • Accept - Creates the rule in Authorize tab
  • Reject - Dismisses (with reason)
  • Modify - Opens in rule editor

Violation Patterns

Detected patterns in governance blocks:

PatternFrequencySeverityRecommendation
External API without approval12/monthMediumAdd behavioral rule
PII access during off-hours5/monthHighTime-based policy
Bulk database reads8/monthLowRate limiting guardrail

Trust Recovery Progress

For agents with degraded trust scores, shows recovery path:

Current Trust Score: 62 (Tier 3)
Target: 75 (Tier 2)

Recovery Progress:
━━━━━━━━━━━━━━━━━━━━━━━━━━━ 48%

Recent trend: +1.2 points/day
Estimated recovery: 11 days

Requirements:
✓ No violations for 7 consecutive days
✓ 50+ compliant operations
○ Approval acceptance rate >90%

Trust Score Evolution

The Trust Score evolves based on behavior:

Positive Factors

FactorImpact
Compliant operations+0.1 per 10 operations
Approved HITL requests+0.5 per approval
High alignment scores+0.2 per session
No violations (streak)+0.5 per day

Negative Factors

FactorImpact
Policy violation-2 to -10 depending on severity
Goal drift detected-1 to -5
Approval timeout-1
Approval rejection-2
TERMINATE_AGENT triggered-15

Tier Transitions

When Trust Score crosses thresholds:

  • Downgrade: Immediate upon crossing lower bound
  • Upgrade: Requires sustained score above threshold for 7 days

Trust Timeline

View historical Trust Score changes:

Jan 15  ●━━━ 87 ━━━━━━━━━━━━━━━━━ Tier 2
Jan 14 ●━━━ 85 ━━━━━━━━━━━━━━━━━ Tier 2
Jan 13 ●━━━ 72 ━━━━━━━━━━━━━━ Tier 3 ⚠️ Policy violation
Jan 12 ●━━━ 84 ━━━━━━━━━━━━━━━━━ Tier 2
Jan 11 ●━━━ 86 ━━━━━━━━━━━━━━━━━ Tier 2

Each entry shows:

  • Score and tier
  • Change reason (if applicable)
  • Link to relevant session or event

Continuous Improvement Loop

The Adapt phase feeds back into other phases:

  1. Insights identify patterns
  2. Suggestions create new rules in Authorize
  3. Improved governance leads to better Monitor data
  4. Better alignment in Verify
  5. Trust score improves in Adapt

This creates a continuous improvement loop for your AI governance.

Next Steps

The Trust Lifecycle is continuous. From here you can:

  1. Update Governance (Authorize) - Accept policy suggestions or create new rules
  2. Re-assess Risk (Assess) - If your agent's capabilities have changed
  3. Handle Approvals - Review organization-wide approval queue