9.2
9.2 Responsible AI, Safety & Model Risk
14 harm taxonomies, fairness and bias metrics, human-in-the-loop escalation, Model & System Cards transparency, and security incident response.
Harm Taxonomy, Fairness/Bias Auditing & Human Oversight
Multi-modal content moderation across 14 harm categories (hate/self-harm/harassment/CBRN), demographic parity and equalized odds fairness auditing, and human-in-the-loop escalation controls.
🏢 Companies
Microsoft (Azure Content Safety)、OpenAI、Anthropic、Google、Arize AI
🛠️ Tech Stack
Harm TaxonomyBias AuditingDemographic ParityEqualized OddsHuman OversightAzure Content SafetyConstitutional AI
💼 Roles & Salary
AI Safety Engineer、Responsible AI Engineer、Safety Researcher
💰 $200K - $430K / year (Responsible AI & Alignment) | ¥550K - ¥1.3M / year
📚 Prerequisites: Harm Taxonomy & Content Moderation Standards • Statistical Fairness Metrics (Demographic Parity) • Constitutional AI Principles & Self-Correction • High-Risk Human Escalation Matrix Design
Model Cards, System Cards Transparency & Incident Response
Standardized Model & System Cards transparency reporting, intended use and out-of-scope boundary definitions, MITRE ATLAS threat mapping, and Coordinated Vulnerability Disclosure (CVD).
🏢 Companies
Hugging Face、OpenAI、Anthropic、Google
🛠️ Tech Stack
Model CardsSystem CardsMITRE ATLASVulnerability DisclosureIncident ResponseHugging Face
💼 Roles & Salary
Responsible AI Engineer、Model Risk Manager、AI Safety Engineer
💰 $190K - $410K / year (Model Risk & Transparency Governance) | ¥520K - ¥1.22M / year
📚 Prerequisites: Hugging Face Model Card YAML Specifications • MITRE ATLAS Threat Matrix for AI Systems • Coordinated Vulnerability Disclosure (CVD) Workflows • Limitations & Out-of-Scope Documentation Standards