Lowered Barriers to Dangerous Knowledge
危險知識的取得門檻降低
Operation & Monitoring
Risk Description
When using a self-hosted version without safeguards, a user can directly obtain structurally complete guidance for dangerous procedures; due to unmitigated control gaps, because it runs in a local environment, no service-side protection or monitoring can intervene, triggering compliance exposure and operational reputational costs.
Framework Mappings
EU AI ActArt.5
NIST AI 600-1CBRN Information or Capabilities、Dangerous, Violent, or Hateful Content
NIST AI RMFMANAGE 2.4
ISO/IEC 5338運作與監控
MIT AI Risk RepositoryDomain 4
Risk Treatment & Implementation Guidance
Build layered safeguards and harmful-content detection at the model layer to block dangerous operational instructions; Assess release strategy and misuse protections for self-hostable versions to reduce bypass of server-side safeguards; Establish tiered controls and refusal policy for high-risk domains