Pillar 2: Management Controls

Ensure operational excellence, risk management, and reliable agent lifecycle management.

Overview

Pillar 2 governs the operational processes required to manage AI agents throughout their lifecycle—from initial development through testing, deployment, monitoring, and eventual retirement. These 24 controls establish the governance framework for change management, model risk, vendor oversight, multi-agent orchestration, customer disclosure, and ongoing supervision required by financial regulators.

Primary Regulatory Alignment: OCC 2011-12 / Fed SR 11-7 (model risk), FINRA 3110 (supervision), SOX 302/404 (internal controls)

Control Categories:

Category	Controls	Focus
Environment Governance	2.1-2.2, 2.15	Managed environments, groups, routing
Lifecycle Management	2.3-2.5, 2.10	Change control, BCDR, testing, patching
Risk Management	2.6-2.8, 2.16-2.18	Model risk, vendor risk, RAG validation, orchestration, conflict testing
Oversight & Compliance	2.9, 2.11-2.14	Performance monitoring, bias testing, supervision, training
Customer Transparency	2.19, 2.22-2.23	AI disclosure, session timeout, consent enforcement
Adversarial & Feature Governance	2.20-2.21, 2.24	Red teaming, marketing claims, feature enablement

Note

Controls 2.6 and 2.11 address model risk management guidance (OCC 2011-12 / SR 11-7) but cover qualitative controls only. Organizations using AI agents for credit decisions or trading should supplement with comprehensive quantitative model validation programs.

Controls

• 2.1 Managed Environments
• 2.2 Environment Groups and Tier Classification
• 2.3 Change Management and Release Planning
• 2.4 Business Continuity and Disaster Recovery
• 2.5 Testing, Validation, and Quality Assurance
• 2.6 Model Risk Management (OCC 2011-12 / SR 11-7)
• 2.7 Vendor and Third-Party Risk Management
• 2.8 Access Control and Segregation of Duties
• 2.9 Agent Performance Monitoring and Optimization
• 2.10 Patch Management and System Updates
• 2.11 Bias Testing and Fairness Assessment
• 2.12 Supervision and Oversight (FINRA Rule 3110)
• 2.13 Documentation and Record Keeping
• 2.14 Training and Awareness Program
• 2.15 Environment Routing and Auto-Provisioning
• 2.16 RAG Source Integrity Validation
• 2.17 Multi-Agent Orchestration Limits
• 2.18 Automated Conflict of Interest Testing
• 2.19 Customer AI Disclosure and Transparency
• 2.20 Adversarial Testing and Red Team Framework
• 2.21 AI Marketing Claims and Substantiation
• 2.22 Inactivity Timeout Enforcement
• 2.23 User Consent and AI Disclosure Enforcement
• 2.24 Agent Feature Enablement and Restriction Governance

---

FSI Agent Governance Framework v1.2 - January 2026