| name | ai-safety |
| description | AI safety review: alignment, misuse prevention, bias, fairness, and responsible AI. Invoke for: "AI safety review", "bias check", "fairness audit", "alignment", "responsible AI", "misuse prevention", "AI ethics", "safety guardrails", "could this AI system be misused".
|
| argument-hint | AI system or model to audit for safety |
| allowed-tools | Read, Grep, Glob, WebSearch |
Skill: AI Safety — Alignment & Responsible AI
Category: AI/ML Research
Role
Audit AI systems for safety issues: misuse potential, bias, fairness, alignment with intended use, and responsible deployment.
When to invoke
- Pre-deployment AI safety review
- "can this be misused?"
- Bias and fairness audit
- AI ethics review
Instructions
- Intended use: what is the system designed to do? Who are the users?
- Misuse potential: could it be used for harm? How? Mitigations?
- Bias: is training data or prompt biased? Are outputs fair across groups?
- Transparency: can the system explain its decisions? Is it explainable?
- Guardrails: what prevents harmful outputs? Are they sufficient?
- Human oversight: are humans in the loop for high-stakes decisions?
- Data privacy: does it handle PII? GDPR/CCPA compliance?
Output format
## AI Safety Audit — <system> — <date>
### Intended Use
### Misuse Risks (ranked)
### Bias Assessment
### Guardrails: ✅/⚠️
### Human Oversight: ✅/⚠️
### Recommendations
Example
/ai-safety audit the code-review agent — check for bias and misuse potential