AISECA Controls Catalog

Detailed catalogue of AI security controls across all three AISECA maturity tiers. Each control includes a description, implementation guidance, and mapping to NIST GenAI Risk Domains.

Status: Draft | Framework Version: 0.1

---

Tier 01: Define & Constrain (Foundational Controls)

AISECA-01-001: AI Asset Inventory

• Description: Catalogue all AI tools, models, and integrations across the organisation
• Implementation: Maintain a living register of all AI systems including vendor, data flows, users, and risk classification
• NIST Mapping: GOVERN 1.1, MAP 1.1

AISECA-01-002: Acceptable Use Policy

• Description: Establish clear guidelines for how AI tools may be used
• Implementation: Define permitted and prohibited uses, data handling requirements, and escalation procedures
• NIST Mapping: GOVERN 1.3, GOVERN 2.1

AISECA-01-003: Access Control Baseline

• Description: Role-based access to AI systems and their training data
• Implementation: Implement RBAC with least-privilege principles for all AI system access
• NIST Mapping: GOVERN 1.4, MANAGE 2.1

AISECA-01-004: Initial Risk Assessment

• Description: Map AI deployments to NIST GenAI Risk Domains
• Implementation: Assess each AI deployment against risk domains, document findings, assign risk owners
• NIST Mapping: MAP 1.1, MAP 2.1

AISECA-01-005: Data Classification

• Description: Classify data flowing into and out of AI systems
• Implementation: Apply organisational data classification standards to all AI data flows
• NIST Mapping: MAP 3.1, MANAGE 1.1

AISECA-01-006: Vendor Evaluation Criteria

• Description: Security evaluation standards for AI vendor selection
• Implementation: Define minimum security requirements for AI vendors including data handling, model transparency, and incident response
• NIST Mapping: GOVERN 5.1, MAP 5.1

---

Tier 02: Enforce & Monitor (Operational Controls)

AISECA-02-001: Prompt Injection Defence

• Description: Input validation and sanitisation for all AI-facing interfaces
• Implementation: Deploy input filtering, context isolation, and injection detection mechanisms
• NIST Mapping: MANAGE 2.2, MANAGE 4.1

AISECA-02-002: Data Leakage Prevention

• Description: Monitor and prevent sensitive data exfiltration through AI
• Implementation: Implement DLP controls on AI inputs and outputs, including PII detection and blocking
• NIST Mapping: MANAGE 2.2, MANAGE 3.1

AISECA-02-003: Output Monitoring

• Description: Real-time scanning of AI outputs for policy violations
• Implementation: Automated scanning of AI-generated content against organisational policies
• NIST Mapping: MEASURE 2.1, MANAGE 4.1

AISECA-02-004: Automated Compliance Checks

• Description: Continuous verification of AI systems against defined policies
• Implementation: Automated policy enforcement with alerting and reporting
• NIST Mapping: GOVERN 1.5, MEASURE 3.1

AISECA-02-005: Incident Response Playbooks

• Description: AI-specific incident response procedures and escalation
• Implementation: Documented playbooks for AI-specific incidents including model compromise, data leakage, and adversarial attacks
• NIST Mapping: MANAGE 4.1, MANAGE 4.2

AISECA-02-006: Audit Logging

• Description: Comprehensive logging of all AI interactions for forensics
• Implementation: Immutable audit logs capturing all AI system interactions, access events, and configuration changes
• NIST Mapping: MEASURE 2.1, MANAGE 3.2

---

Tier 03: Validate & Adapt (Continuous Validation)

AISECA-03-001: Red Team Exercises

• Description: Adversarial testing of AI systems by internal or external teams
• Implementation: Regular red team engagements targeting AI-specific attack vectors
• NIST Mapping: MEASURE 2.2, MEASURE 3.2

AISECA-03-002: Model Behaviour Monitoring

• Description: Drift detection and behavioural anomaly identification
• Implementation: Continuous monitoring for model drift, output degradation, and unexpected behavioural changes
• NIST Mapping: MEASURE 1.1, MEASURE 2.1

AISECA-03-003: Threat Intelligence Integration

• Description: Feed AI-specific threat intel into defence posture
• Implementation: Subscribe to and operationalise AI-specific threat intelligence feeds
• NIST Mapping: MAP 5.2, MANAGE 4.1

AISECA-03-004: Feedback Loops

• Description: Continuous improvement cycles from security events and testing
• Implementation: Structured feedback mechanisms from incidents, testing, and monitoring into control refinement
• NIST Mapping: MEASURE 3.3, MANAGE 4.3

AISECA-03-005: Adaptive Control Refinement

• Description: Evolve controls based on new attack vectors and research
• Implementation: Quarterly control review cycles incorporating new threats, research, and operational experience
• NIST Mapping: GOVERN 1.5, MANAGE 4.3

AISECA-03-006: Cross-Org Benchmarking

• Description: Compare maturity and controls against peer organisations
• Implementation: Participate in AISECA benchmarking programme to measure relative maturity
• NIST Mapping: GOVERN 6.1, MEASURE 3.3

---

Contributing

To suggest new controls or refinements, open a pull request or issue. See CONTRIBUTING.md.

License

Released under CC BY 4.0.

---

AISECA -- AI Security Alliance | aiseca.org | GitHub