AI Safety Risk Taxonomy
An AI Safety Risk Taxonomy is a risk classification framework that categorizes AI safety threats, failure modes, and harmful outcomes.
- AKA: AI Risk Classification System, AI Safety Threat Framework, AI Hazard Taxonomy, AI Risk Categorization.
- Context:
- It can typically organize Risk Types by severity, likelihood, and mitigation difficultys.
- It can typically distinguish Technical Risks from misuse risks and structural risks.
- It can typically guide Risk Assessments and safety research prioritys.
- It can typically evolve with New Capabilitys and emerging threats.
- It can often inform Safety Standards and regulatory frameworks.
- It can often facilitate Risk Communication between stakeholders.
- It can often enable Systematic Defenses against categorized threats.
- It can range from being a Simple Risk Taxonomy to being a Comprehensive Risk Taxonomy, depending on its coverage depth.
- It can range from being a Static Risk Taxonomy to being a Dynamic Risk Taxonomy, depending on its update frequency.
- It can range from being a Technical Risk Taxonomy to being a Sociotechnical Risk Taxonomy, depending on its scope breadth.
- It can range from being a Academic Risk Taxonomy to being a Operational Risk Taxonomy, depending on its application context.
- ...
- Example:
- Capability Risk Categorys, such as:
- Deception Risks including AI Deceptive Behaviors.
- Power-Seeking Risks from instrumental goals.
- Misalignment Risks between AI objectives and human values.
- Security Risk Categorys, such as:
- AI Security Risks threatening model integritys.
- Cyberweapon Risks from AI-enhanced attacks.
- Biosecurity Risks from AI-designed pathogens.
- Systemic Risk Categorys, such as:
- ...
- Capability Risk Categorys, such as:
- Counter-Example:
- General Risk Framework, which lacks AI specificity.
- Threat Model, which focuses on specific attacks not categories.
- Safety Checklist, which lists requirements not risk types.
- Incident Database, which records actual events not potential risks.
- See: AI Safety, Risk Taxonomy, AI Security Risk, AI Deceptive Behavior, Existential Risk, AI Alignment Problem, Safety Assessment, Risk Management, Threat Modeling, AI Governance Framework.