# SP-045 AI Governance and Responsible AI

Status: active. Release 2.0. Modified 2026-02-13. Licence: CC BY-SA 4.0.

Scope: A governance pattern for establishing and operating an AI management system (AIMS) across the enterprise. Covers training data governance, model lifecycle management, transparency and explainability, AI-specific risk and impact assessment, human oversight, and responsible AI monitoring.
Use when: Organisation deploys AI systems that make or support decisions affecting individuals (credit, employment, insurance, healthcare, law enforcement). EU AI Act applies to the organisation's AI systems.
Not when: Organisation uses AI only for internal, non-consequential tasks (e.g., code completion, document formatting) with no impact on individuals.

## Controls (34, NIST SP 800-53 ids)
- Critical (10): AU-02, AU-03, CA-07, CM-02, PM-09, PT-04, RA-03, RA-08, SA-11, SI-10
- Important (21): AC-05, AC-06, AT-02, AT-03, AU-06, AU-12, CA-02, CM-03, CM-04, PM-14, PM-25, PM-28, PM-32, PT-02, PT-03, PT-05, RA-07, RA-09, SA-03, SA-08, SI-04
- Standard (3): PL-02, PS-06, SA-10

## What each critical control mitigates (10)
- AU-02 Event Logging: T-AIG-002, T-AIG-010
- AU-03 Content of Audit Records: T-AIG-002
- CA-07 Continuous Monitoring: T-AIG-001, T-AIG-005
- CM-02 Baseline Configuration: T-AIG-003, T-AIG-008
- PM-09 Risk Management Strategy: T-AIG-011
- PT-04 Consent: T-AIG-006
- RA-03 Risk Assessment: T-AIG-011
- RA-08 Privacy Impact Assessments: T-AIG-004, T-AIG-011, T-AIG-012
- SA-11 Developer Testing and Evaluation: T-AIG-001
- SI-10 Information Input Validation: T-AIG-001, T-AIG-003

## Threats and the controls that mitigate them (11)
- T-AIG-001 Training data bias producing discriminatory AI outputs across protected characteristics: SI-10, SA-11, CA-07
- T-AIG-002 Opaque AI decisions in regulated contexts without explainability or contestability: PT-05, AU-02, AU-03
- T-AIG-003 Training data poisoning corrupting model behaviour at scale: SI-10, SA-03, CM-02
- T-AIG-004 Uncontrolled model deployment without impact assessment or approval gate: RA-08, CM-03, AC-05
- T-AIG-005 Concept drift degrading model accuracy and fairness over time: CA-07, SI-04, CM-04
- T-AIG-006 Training data containing PII processed without lawful basis or consent: PT-02, PT-04, PM-25
- T-AIG-008 Model versioning failure enabling rollback to unsafe or biased version: CM-02, SA-10, CM-03
- T-AIG-009 Insufficient human oversight of high-risk autonomous AI decisions: AC-05, AT-02, PS-06
- T-AIG-010 Failure to disclose AI involvement to affected individuals: PT-05, PT-03, AU-02
- T-AIG-011 Regulatory non-compliance with EU AI Act or equivalent AI legislation: RA-08, RA-03, PM-09
- T-AIG-012 AI system causing unassessed societal or environmental harm: RA-08, RA-09, CA-02

## More
- The critical controls and what each mitigates, as JSON (a few KB): /api/v1/patterns/SP-045/crosswalk?emphasis=critical
- The same for every control, with its clauses in a framework: /api/v1/patterns/SP-045/crosswalk?framework={framework id}. Framework ids are listed in /llms.txt
- The pattern's prose, examples and references as JSON, 30 KB: /api/v1/patterns/SP-045
- Page for people: /patterns/sp-045/
- Found an error? Open an issue at https://github.com/opensecurityarchitecture/osa-data/issues with the id, what OSA says and what the source says.
- Related: SP-027 Secure LLM Usage; SP-047 Secure Agentic AI Frameworks; SP-013 Data Security Pattern; SP-012 Secure Software Development Lifecycle; SP-042 Third Party Risk Management; SP-043 Security Metrics and Measurement; SP-028 Secure DevOps Pipeline

This card, the API and the page are generated from one file. Checking one against another adds no evidence.