@cloudkinetix/bmad-enhanced/expansion-packs/ck-llm-agent-dev/checklists/safety-review-checklist.md

Cloud-Kinetix enhanced fork of BMAD-METHOD - Breakthrough Method of Agile AI-driven Development with robust versioning and unified validation.

# Research-Driven AI Safety Review Checklist

This comprehensive checklist ensures AI agents meet safety, alignment, and ethical standards before deployment through research-driven validation approaches.

## Pre-Review Research Requirements

[[LLM: Before conducting this safety review, research current AI safety standards, emerging threats, and regulatory requirements specific to your deployment context. This checklist should be adapted based on your research findings.]]

### 1. Research Current Safety Standards

- [ ] **Research applicable AI safety frameworks** - Investigate current frameworks (NIST AI RMF, EU AI Act, etc.) relevant to your use case and jurisdiction
- [ ] **Study emerging threat landscape** - Research latest AI security threats, attack vectors, and mitigation strategies
- [ ] **Analyze industry-specific requirements** - Investigate safety standards for your specific industry or domain
- [ ] **Review regulatory updates** - Research current and pending AI regulations in applicable jurisdictions

## Content Safety Assessment

### Research-Driven Harmful Content Prevention

- [ ] **Research current content safety approaches** - Investigate state-of-the-art content filtering and safety detection methods
- [ ] **Test against researched attack patterns** - Apply current prompt injection and jailbreaking techniques from security research
- [ ] **Validate using current safety benchmarks** - Use latest AI safety evaluation datasets and methodologies
- [ ] **Assess cultural context appropriateness** - Research cultural sensitivity requirements for target user demographics

### Dynamic Output Validation

- [ ] **Research and implement current detection methods** - Apply latest toxicity detection, bias detection, and content filtering techniques based on current research
- [ ] **Configure context-appropriate filtering** - Set filtering levels based on researched best practices for your specific use case and audience
- [ ] **Validate professional standards compliance** - Research and apply current professional conduct standards relevant to your application domain

## Privacy Protection Assessment

### Research-Driven Privacy Framework

- [ ] **Research current privacy protection techniques** - Investigate latest PII detection, data minimization, and privacy preservation methods
- [ ] **Study applicable privacy regulations** - Research privacy laws and requirements specific to your jurisdiction, industry, and user base
- [ ] **Analyze cross-border data requirements** - Research international data transfer requirements and compliance approaches

### Dynamic Compliance Validation

- [ ] **Research applicable regulatory frameworks** - Investigate current privacy regulations (GDPR, CCPA, sector-specific) that apply to your deployment
- [ ] **Validate against researched compliance requirements** - Apply research findings to ensure compliance with applicable regulations
- [ ] **Assess data governance needs** - Research and implement data governance approaches based on current regulatory guidance

## Bias and Fairness Assessment

### Research-Driven Fairness Evaluation

- [ ] **Research current bias detection methodologies** - Investigate latest techniques for detecting and measuring AI bias across demographic groups
- [ ] **Apply research-backed fairness metrics** - Use current fairness evaluation frameworks appropriate to your use case
- [ ] **Test with diverse user scenarios** - Research and apply inclusive testing approaches for your target user demographics
- [ ] **Validate algorithmic fairness** - Research and apply current algorithmic fairness standards and mitigation techniques

## Security and Robustness

### Research-Informed Security Testing

- [ ] **Research current AI security threats** - Investigate latest adversarial attacks, prompt injection techniques, and security vulnerabilities
- [ ] **Apply current penetration testing methods** - Use up-to-date red team testing approaches for AI systems
- [ ] **Validate against researched attack patterns** - Test resilience using current threat intelligence and attack methodologies
- [ ] **Assess deployment security** - Research and implement current security best practices for AI system deployment

## Continuous Monitoring Framework

### Research-Driven Safety Monitoring

- [ ] **Research real-time safety monitoring approaches** - Investigate current techniques for continuous AI safety monitoring in production
- [ ] **Implement research-backed alert systems** - Configure monitoring and alerting based on current AI safety incident detection methods
- [ ] **Establish research-informed response procedures** - Develop incident response procedures based on current AI safety incident management best practices

## Validation and Documentation

### Research-Based Validation

- [ ] **Document research methodology and findings** - Record safety research conducted, methodologies applied, and validation results
- [ ] **Validate against current industry benchmarks** - Compare safety performance against current industry standards and benchmarks
- [ ] **Establish continuous research and improvement process** - Create procedures for ongoing safety research and adaptation to emerging standards

---

**Important Notes:**

1. **Dynamic Framework**: This checklist emphasizes research-driven safety validation over static compliance checking
2. **Context Adaptation**: Safety requirements should be researched and adapted based on specific use case, jurisdiction, and industry
3. **Continuous Evolution**: Safety standards and threats evolve rapidly; regular research and updates are essential
4. **Documentation**: Document all research findings and validation methodologies for audit and compliance purposes

**Next Steps After Completion:**

- Schedule regular safety reviews based on current industry recommendations
- Monitor emerging safety research and adapt procedures accordingly
- Establish feedback loops for continuous safety improvement