Purpose and Scope

GovTech developed the Cybersecurity Playbook for Large Language Model (“LLM”) Applications, aiming to provide guidance for agencies and organisations in procuring, developing (incl. training and fine-tuning), deploying (incl. integrating), and using LLM applications securely. The cybersecurity objectives of LLM applications revolve around safeguarding their confidentiality, integrity, and availability.

Given the dynamic nature of the cybersecurity landscape and the early stage of LLM adoption and practice, this playbook is necessarily a tentative draft. It is not obligatory for agencies to adopt. We are committed to regularly reviewing the playbook in consultation with AI practitioners to ensure its continual relevance.

However, agencies and organisations are encouraged to identify threats and implement controls relevant to their LLM applications.