AI Risk Families
AI risk taxonomy
AI Risk Families
A high-level categorization of ways that capable AI systems can create harm, each grounded in recognizable scenes from pop culture.
05 records
- 01
Accidents
Risks from AI systems causing unintended harm when they fail, behave unpredictably, or produce unreliable results.
96 scenarios - 02
Malicious use
Risks from people deliberately using AI to cause harm.
101 scenarios - 03
Misalignment
Risks from AI systems pursuing goals or behaving in ways that conflict with human intentions or values.
143 scenarios - 04
Security
Risks from weak security or oversight leaving AI systems vulnerable to hacking, theft, manipulation, or irresponsible use.
181 scenarios - 05
Systemic
Risks from widespread AI adoption concentrating power, deepening inequality, or causing failures that spread across society.
155 scenarios