Skip to main content
A Singapore Government Agency Website How to identify
Official website links end with .gov.sg
Secure websites use HTTPS
Look for a lock () or https:// as an added precaution. Share sensitive information only on official, secure websites.
Government officials will never ask you to transfer money or disclose bank log-in details over a phone call.

Call the 24/7 ScamShield Helpline at 1799 if you are unsure if something is a scam.

Cyber Security Agency of Singapore
Publications

Security Assessment Handbook for Agentic AI Coding Assistants

15 October 2026

CSA releases a Security Assessment Handbook to support security practitioners and developers adopt agentic AI coding assistants more securely.

With the increasing adoption of AI across sectors and the greater autonomy of agentic AI capabilities (e.g., in formulating plans and taking actions independently), ensuring that AI systems are secure and trustworthy becomes ever more critical. To support secure adoption of AI:

  • CSA had released the Guidelines and Companion Guide on Securing AI Systems in 2024, which set out foundational AI security principles and controls across the AI lifecycle.  

  • CSA had further built on the Guidelines and Companion Guide to release the Securing Agentic AI in 2025, which introduced additional security considerations arising from the increased autonomy, access and reach of agentic AI systems. 

AI coding assistants is a common use case of agentic AI, given their rapid adoption across enterprises and the level of autonomy and access they may be granted within software development environments. These capabilities introduce security risks, where threat actors could exploit and manipulate AI coding assistants into executing malicious commands, exfiltrating sensitive information, or abusing their access to development environments and external tools. 

Having performed security assessments on AI coding assistants, CSA developed the Security Assessment Handbook for Coding Assistants to provide security practitioners and developers with a practical methodology to assess their security and determine appropriate safeguards for their deployments. Specifically, the handbook:

  1. Explains the general workflow of how IDE-based and CLI-based AI coding assistants perform software development tasks and interface with external tools and information 

  1. Provides trust boundaries, threats and attack paths, with mapped tactics and techniques from MITRE ATLAS and other resources 

  1. Provides practical controls mapped to each threat and trust boundary, drawn from the CSA Addendum on Securing Agentic AI (opens in new tab). The mitigations are tiered based on criticality, so that they can be prioritised for implementation, if needed. 

The Security Assessment Handbook is designed as an interactive resource to help users better understand how attacks could unfold across an AI coding assistant environment. Rather than presenting attack vectors and mitigations in isolation, users can follow the relevant attack paths across different components and trust boundaries, and see how individual risks may be introduced, propagated and mitigated. This provides a more intuitive way for security practitioners and developers to understand the relationships between the system architecture, potential threats and corresponding security controls: