AI Governance Map

Using an LLM pipeline, we rated each document in CSET Emerging Technology Observatory's AGORA dataset against the MIT AI Risk Domain Taxonomy. See the pilot blog post for discussion of the methodology.

Insights from Risk Subdomain Analysis

• The risk subdomains that are covered by the highest number of documents in the AGORA dataset are: AI System Security Vulnerabilities and Attacks, Governance Failure, Lack of Capability or Robustness, Compromise of Privacy, Lack of Transparency
• The risk subdomains covered by the fewest documents in the AGORA dataset are: AI Welfare and Rights, Multi-agent risks, Economic and Cultural Devaluation, Power Centralization, Environmental Harm
• Minority vs majority focus of subdomains within AGORA dataset: The most-covered subdomains tend to focus on model safety and established regulatory concerns (security, privacy, transparency), whilst less-well covered subdomains include socioeconomic risks (economic devaluation, power centralization) and emerging considerations (multi-agent risks, AI welfare).
• These coverage patterns may reflect the relative maturity of topics in policy discourse, the availability of technical or legal frameworks to address them, or how directly risks map to established regulatory concepts.
• The patterns observed here reflect the composition of the AGORA dataset only, and may not generalize to AI governance discourse as a whole.