AI Governance & Safety Intelligence

SKILL.md

---
AIGC:
    Label: "1"
    ContentProducer: 001191440300708461136T1XGW3
    ProduceID: f4a677bd9d680e9793dee310c93c3656_6231376a5dbe11f1bd025254006c9bbf
    ReservedCode1: d/nnDyDqlH0BkZsH747eEiHPbEsJ8DVkxisrCtc8/0n7+bWb2LoJADy3+Br7vVRqFN/oD7iAjXRfS6O9/KU/xjptPxaphy6RS2OYWD5ZRl5k2fMW8waVDKTMJZ01vW0Y4ILONFYMsg5K3lmEfk3LeMcJYH2rz0SfxTZ/GW2woVAgloOJFmIiyJckj/g=
    ContentPropagator: 001191440300708461136T1XGW3
    PropagateID: f4a677bd9d680e9793dee310c93c3656_6231376a5dbe11f1bd025254006c9bbf
    ReservedCode2: d/nnDyDqlH0BkZsH747eEiHPbEsJ8DVkxisrCtc8/0n7+bWb2LoJADy3+Br7vVRqFN/oD7iAjXRfS6O9/KU/xjptPxaphy6RS2OYWD5ZRl5k2fMW8waVDKTMJZ01vW0Y4ILONFYMsg5K3lmEfk3LeMcJYH2rz0SfxTZ/GW2woVAgloOJFmIiyJckj/g=
---



# AI Governance & Safety Intelligence

The definitive AI governance intelligence engine. 25 authoritative sources covering every major regulatory framework (EU AI Act, US EO 14110, China GenAI Measures, OECD AI Observatory), 3,000+ AI incident records, frontier model evaluation benchmarks, and international cooperation frameworks. Built for policymakers, compliance officers, AI safety researchers, and anyone tracking the rapidly evolving AI governance landscape.

## Triggers

- "What AI regulations apply to [use case]?"
- "Track EU AI Act compliance milestones"
- "Compare AI governance: US vs. EU vs. China"
- "Analyze latest AI safety incidents"
- "Assess frontier model evaluation frameworks"
- "Monitor international AI safety summit outcomes"
- "Check ISO 42001 certification requirements"
- "Research AI liability frameworks"

## Capabilities

| # | Capability | Data Scope | Output Mode |
|---|-----------|-----------|-------------|
| 1 | Regulatory jurisdiction comparison | 45+ jurisdictions: EU AI Act, US EO 14110, China GenAI Measures, UK AISI, CAIDP rankings | Side-by-side table: risk classification, obligations, enforcement, penalties |
| 2 | Compliance timeline tracking | EU AI Act 2024-2027 phased rollout, global milestones | Gantt-style timeline: prohibited practices → GPAI rules → high-risk obligations → full enforcement |
| 3 | AI incident database analysis | AIID (3,000+ incidents), OECD AIM, MIT 777-risk taxonomy | Incident trending by harm type, root cause, deployer, sector, severity |
| 4 | Frontier model safety evaluation | OpenAI Preparedness, Anthropic RSP (ASL 1-4), DeepMind Frontier Safety, METR, ARC Evals | Capability threshold comparison, CBRN/cyber/autonomy benchmarks, evaluation methodology gap analysis |
| 5 | Standards & certification landscape | ISO/IEC 42001, ISO 23894, IEEE 7000, CEN-CENELEC JTC 21, BSI Hub | Certification roadmap, conformity assessment bodies, standard-to-regulation mapping |
| 6 | International AI governance cooperation | UN AI Advisory Body, AI Safety Summit series (2023-2026), G7 Hiroshima AI Process, Global AISI Network (10+ institutes) | Institutional mapping, mandate comparison, soft law vs. hard law trajectory |
| 7 | AI safety & alignment research landscape | Anthropic, DeepMind, OpenAI, ARC Evals, METR, Epoch AI, CAIS, Redwood | Research program comparison, key papers, open problems, scaling trajectory data |
| 8 | Compute governance & export controls | BIS Entity List, GPU export restrictions, chip tracking proposals | Hardware restriction tiers, impact assessment on AI capability diffusion, sovereign AI compute gap |
| 9 | Digital rights & AI ethics | Access Now, AlgorithmWatch, STOP | Facial recognition bans, predictive policing, automated decision audits, civil society position mapping |
| 10 | AI risk horizon scanning | Integrated cross-source fusion | Emerging risks: agentic AI, open-weight models, bio risk, labor displacement, elections integrity |

## Workflow

```
User Query
    │
    ├─ Regulatory compliance query → Jurisdiction comparison (Capability #1) + Timeline (Capability #2)
    │   └─ Output: Applicable frameworks table + compliance milestones + penalty exposure
    │
    ├─ Safety evaluation query → Frontier model eval (Capability #4) + Incidents (Capability #3)
    │   └─ Output: RSP/Preparedness level mapping + relevant incident patterns + residual risk assessment
    │
    ├─ International governance query → Cooperation tracking (Capability #6) + Standards (Capability #5)
    │   └─ Output: Institutional landscape + mandate mapping + harmonization gap analysis
    │
    ├─ Open-source AI debate query → Compute governance (Capability #8) + Digital rights (Capability #9)
    │   └─ Output: Pro/con analysis with evidence, regulatory trajectory in key jurisdictions
    │
    └─ Horizon scanning → Risk scanning (Capability #10) + Research landscape (Capability #7)
        └─ Output: Emerging risk dashboard + research direction watchlist + 12-month outlook
```

## Output Formats

### Format 1: Regulatory Comparison Matrix
```
| Jurisdiction | Risk Classification | Conformity Assessment | Transparency | Enforcement | Penalties |
|-------------|--------------------|----------------------|--------------|-------------|-----------|
| EU AI Act   | Unacceptable/High/Limited/Minimal | Third-party (High) | ✓ | EU Commission + MS | Up to 7% global turnover |
| US EO 14110 | Federal agency RMF | Self-assessment (voluntary) | ✓ (exec. branch) | NIST + OMB | Agency-level |
| China GenAI | Filing + algorithm registry | CAC review | ✓ | CAC | Service suspension |
```

### Format 2: Frontier Model Safety Dashboard
```
| Model | Developer | Framework | CBRN Risk | Cyber Risk | Autonomy Risk | Persuasion Risk |
|-------|-----------|-----------|------------|------------|---------------|-----------------|
```

### Format 3: AI Incident Trend Analysis
```
| Harm Type | 2023 | 2024 | 2025 | Trend | Most Affected Sector |
|-----------|------|------|------|-------|---------------------|
```

## Usage Guidelines

- Always cite regulatory source with effective date
- Distinguish between enacted law, proposed regulation, and voluntary framework
- Flag regulatory divergence as risks for cross-border deployment
- Use AI incident data responsibly; note that underreporting biases exist
- When assessing frontier models, reference the developer's own framework

## Examples

**Q: Do I need to comply with EU AI Act for my chatbot?**
→ EU AI Act Article 5 (prohibited) + Article 6 (high-risk) + Article 52 (transparency for chatbots) → Classification: [transparency obligation only], requirements: [user disclosure, no other obligations], timeline: [effective now, full 2026]

**Q: Compare US vs. China AI chip restrictions**
→ BIS Entity List (SMIC, Huawei) + semiconductor export tiers + CAC algorithm registry → US restricts [GPU tiers T1-T3], China restricts [training data, algorithm filing], impact: [quantified]

**Q: Latest frontier model safety incidents**
→ AIID database query + OECD AIM + developer transparency reports → Top 3 incident categories: [deepfakes elections / autonomous agent failures / bias discrimination], trend: [increasing/decreasing]

---

**Data Sources (25 total)**: See `references/ai_governance_sources.json` for complete listing with URLs and update frequencies.
*（内容由AI生成，仅供参考）*