The AI risk repository: A meta-review, database, and taxonomy of risks from artificial intelligence
Researchers have developed a unified database to standardize how we describe and manage risks in artificial intelligence.
The project identifies 1,725 specific risks collected from 74 different existing frameworks. It categorizes these risks using two separate systems to ensure clarity. The Causal Taxonomy looks at the origin of a risk. It asks if the risk comes from a human or an AI, if it was intentional, and if it happens before or after a system is deployed. The Domain Taxonomy focuses on the actual effects of the technology. It covers seven key areas, including privacy violations, discrimination, misinformation, and the development of weapons.
This work is important because different sectors often use different terms for the same problems. When academics, policymakers, and tech companies do not speak the same language, it is difficult to create coordinated safety measures. For those in the public sector, this repository provides a structured way to build regulations and audit AI systems. It helps move from abstract concerns to a concrete framework that works for both engineers and policy makers. Having a shared reference point makes it easier to ensure that AI adoption remains safe and accountable across different countries and industries.
How will a common language for risk change your approach to AI policy?
#AIRisk #AIGovernance #PublicSector #ArtificialIntelligence #TechPolicy
https://doi.org/10.1016/j.patter.2026.101517