Compliance Checklists for Agentic AI

Regulatory Framework Templates

Document ID: F5-COMPLIANCE-CHECKLISTS
Version: 1.0
Category: Governance

---

SOC 2 Type II Checklist

CC6: Logical and Physical Access Controls

Control	Requirement	Agentic Implementation	Status
CC6.1	Access control policies	Agent authentication via API keys	☐
CC6.2	User registration/deregistration	Session management, token rotation	☐
CC6.3	User authentication	OAuth 2.0 / API key validation	☐
CC6.6	Restriction of privileged access	Tool permission scoping	☐
CC6.7	Access review	Agent capability audits	☐

CC7: System Operations

Control	Requirement	Agentic Implementation	Status
CC7.1	Vulnerability management	LLM input sanitization	☐
CC7.2	Change management	Agent versioning, deployment gates	☐
CC7.3	Security event monitoring	Action logging, anomaly detection	☐
CC7.4	Incident response	Agent fallback procedures	☐

CC8: Change Management

Control	Requirement	Agentic Implementation	Status
CC8.1	Infrastructure changes	Agent deployment pipelines	☐

Agentic-Specific Controls

Control	Description	Implementation
AGT-01	Audit trail for all agent actions	VE paradigm with immutable logging
AGT-02	Human-in-the-loop for sensitive operations	Confirmation gates in protocols
AGT-03	Output validation and sanitization	Schema validation, PII filtering
AGT-04	Token budget controls	Per-task and per-session limits

---

HIPAA Checklist

Administrative Safeguards (§164.308)

Requirement	Agentic Implementation	Status
Risk analysis	Threat modeling for agent access	☐
Workforce training	Operator training on agent supervision	☐
Access management	Role-based agent permissions	☐
Contingency plan	Agent failover procedures	☐

Technical Safeguards (§164.312)

Requirement	Agentic Implementation	Status
Access control	Agent authentication, session management	☐
Audit controls	VE paradigm audit logging	☐
Integrity controls	Input/output validation	☐
Transmission security	TLS 1.3 for all communications	☐

PHI Handling Requirements

REQUIRED CONTROLS FOR AGENTIC PHI ACCESS:

1. MINIMUM NECESSARY
   - Agent retrieves only required PHI
   - Context window excludes unnecessary PHI
   - Tool outputs filtered for PHI

2. LOGGING
   - Every PHI access logged with:
     * Timestamp
     * Data elements accessed
     * Purpose
     * Agent/user identifier

3. ENCRYPTION
   - PHI encrypted in transit (TLS 1.3)
   - PHI encrypted at rest (AES-256)
   - PHI masked in logs: SSN → XXX-XX-1234

4. BAA REQUIREMENTS
   - LLM provider BAA in place
   - Cloud provider BAA in place
   - Vector database BAA if storing PHI

---

GDPR Checklist

Article 5: Data Processing Principles

Principle	Agentic Implementation	Status
Lawfulness	Document legal basis for agent processing	☐
Purpose limitation	Scope agent access to stated purposes	☐
Data minimization	Retrieve only necessary data	☐
Accuracy	Validate agent outputs, correction mechanisms	☐
Storage limitation	Implement retention policies for agent data	☐
Integrity & confidentiality	Encryption, access controls	☐

Article 17: Right to Erasure

IMPLEMENTATION REQUIREMENTS:

1. ERASURE SCOPE
   - User data in knowledge bases
   - Episodic memory entries
   - Audit logs (after retention period)
   - Training data (if applicable)

2. VERIFICATION
   - Confirm deletion across all stores
   - Vector embeddings removed
   - Backup purging scheduled

3. RESPONSE TIME
   - 30 days maximum
   - Automated pipeline recommended

Article 22: Automated Decision-Making

Requirement	Implementation	Status
Right to human intervention	Escalation paths defined	☐
Right to explanation	Decision rationale logging	☐
Right to contest	Appeal process documented	☐

---

SOX Compliance (Section 404)

Internal Controls for Agentic Financial Processes

Control	Description	Implementation
Segregation of duties	Agent cannot approve own outputs	Multi-agent with human approval
Authorization	Defined approval thresholds	VE protocol gates
Audit trail	Complete action history	Immutable audit logging
Reconciliation	Agent outputs verified	Automated validation

Required Documentation

• Agent process flowcharts
• Control matrices for each agent workflow
• Risk assessments for automated decisions
• Testing evidence (quarterly)
• Management certification

---

Financial Services (SEC/FINRA)

Recordkeeping Requirements

RETENTION REQUIREMENTS:

Books & Records (17a-4):
- Agent communications: 3 years (first 2 accessible)
- Transaction records: 6 years
- Audit logs: 6 years

Format:
- Write-once, read-many (WORM) compliant
- Indexed and searchable
- Reproducible in original format

Supervision Requirements

Requirement	Agentic Implementation
Written procedures	Documented agent protocols
Review of activities	Human review of agent outputs
Annual compliance review	Agent audit program
Training	Operator certification

---

AI-Specific Regulations

EU AI Act Readiness

Requirement	High-Risk AI	Limited Risk
Risk assessment	Required	Recommended
Data governance	Required	Recommended
Technical documentation	Required	Basic
Transparency	Full disclosure	Notification
Human oversight	Mandatory	Optional
Accuracy/robustness	Testing required	Best effort

Recommended Controls for All AI

BASELINE AI GOVERNANCE:

1. TRANSPARENCY
   - Disclose AI use to users
   - Explain capabilities and limitations
   - Provide human alternative option

2. BIAS MONITORING
   - Regular output audits
   - Demographic fairness testing
   - Correction mechanisms

3. SAFETY
   - Content filtering
   - Harmful output detection
   - Incident reporting

4. ACCOUNTABILITY
   - Clear ownership
   - Escalation paths
   - Regular reviews

---

Compliance Quick Reference

Regulation	Key Agent Requirement	Recommended Paradigm
SOC 2	Audit logging, access control	VE
HIPAA	PHI protection, minimum necessary	VE + GS
GDPR	Data minimization, erasure	VE
SOX	Segregation, authorization	VE + Human approval
SEC/FINRA	Recordkeeping, supervision	VE
EU AI Act	Transparency, human oversight	VE + Human oversight

---

Document maintained by CODITECT Compliance Team. Consult legal counsel for specific implementations.