Asia AI Safety AtlasStatic, source-backed stakeholder intelligence

Methodology

Evidence standards for the atlas

The atlas is built as a static-first research product. The data model stays small and explicit until the source workflow is reliable enough for a database or CMS.

Evidence rules

  • Every public fact should trace to a source record.
  • Research briefs are discovery inputs, not final authority.
  • For non-Western actors, local-language research comes first.
  • Relationship edges require direct evidence, not inferred proximity.
  • Confidence ratings and open questions are preferred over overclaiming.

Controlled focus areas

Frontier Ai SafetyTechnical AlignmentAi GovernanceStandardsEvaluationModel SecurityCybersecurityBiosecurityContent SafetyEthicsOpen Source GovernanceInternational DialogueCapacity BuildingAi AssuranceRed TeamingAgent SafetyDangerous Capability EvaluationOpen Model GovernancePolicy RegulationTechnical Safety ResearchTesting InfrastructureDialogue ChannelFrontier Model Developer

Conservative relationship labels

OverseesHostsMember OfPublishesCo IssuesMaintainsEstablishesChairsOperatesOrganizesCo HostsParticipates InConducts Testing WithSupplementsAdoptsEvaluatesCertifiesAffiliated With

The atlas avoids partnership wording unless a source explicitly supports it. Broad proximity is captured with open questions or lower-confidence notes instead of stronger edge labels.