Skip to content

Implementating the ARC Framework

A well-known adage in the Singapore civil service is "Policy is implementation and implementation is policy",1 and this is resoundingly true for AI governance. The ARC Framework aims to bridge the gap between academic work on agentic AI safety and security with real-world governance considerations by synthesising a systematic way to:

  • Anticipate safety and security risks before they manifest
  • Scale governance across diverse AI applications
  • Maintain consistency while allowing innovation
  • Focus human oversight on critical decisions
  • Adapt to technological change without rebuilding governance

Click here for a downloadable version of the Agentic Risk & Capability Framework - we provide the taxonomy, risk mapping, and technical controls in Google Sheets so that it's easy to download and adapt it.

Here, we outline how two different parties, AI developers and governance teams, can apply the ARC Framework for their own use.

  • For governance teams, the ARC Framework provides a scalable and sustainable way to manage the immediate and emergent risks of agentic AI systems in their organisation.
  • For AI developers, the ARC Framework helps in identifying the relevant risks and controls to mitigate the safety and security risks for their agentic AI system.

  1. Ho, Peter. Opening Address at 2010 Administrative Service Dinner and Promotion Ceremony. Public Service Division, March 2010. https://www.psd.gov.sg/files/opening-address-by-mr-peter-ho-at-2010-administrative-service-dinner-and-promotion-ceremony.pdf. Accessed: 2025-07-21. ↩