Back to Policy Library 🏢 Sector Regulation 2026-06

Policy / Project Profile

Securing Agentic AI — Addendum to the Guidelines on Securing AI Systems

The agentic-AI addendum to CSA’s 2024 Guidelines on Securing AI Systems: draft opened for consultation on 22 October 2025, finalised version released on 17 June 2026.

Category
Sector Regulation
Published / Updated
2026-06
Issuing body
Cyber Security Agency (CSA)

Detailed Notes

CSA’s Guidelines and Companion Guide on Securing AI Systems, published in October 2024, address AI systems in general. Agentic AI understands context and acts on its own, giving it a different attack surface, so CSA issued a separate addendum for it.

The timeline runs in two stages, which are easy to conflate: the draft was published for public consultation on 22 October 2025, closing 31 December 2025; the finalised version was released on 17 June 2026.

The addendum covers three things: identifying and assessing risk according to agent capabilities, including mapping workflows to surface exploitable weak points; practical controls across the development lifecycle, with examples for different levels of autonomy; and worked scenarios such as app development assistants, automated client onboarding, and automated fraud detection.

Resources

More in This Category

Cite this record

Singapore AI Observatory. Securing Agentic AI — Addendum to the Guidelines on Securing AI Systems. Retrieved 2026-08-20, https://sgai.md/policies/securing-agentic-ai-addendum/

More on these topics