Artificial Intelligence Policy Tracker
Search
Menu

AI policies in Singapore

Search and filter source-backed AI laws, regulations, standards, guidance and consultations. Each record shows its status, key dates, verification state and official source.

Results

5 of 244 instruments match

Singapore Framework Voluntary standard

Singapore agentic AI governance framework

Model AI Governance Framework for Agentic AI

A governance framework for AI agents that reason and act on a user's behalf, in four areas: assessing and bounding risks upfront, making humans meaningfully accountable, technical controls across the agent lifecycle, and enabling end-user responsibility. IMDA treats it as a living document and has since added case studies and input from more than 50 organisations.

Pending review (a factual check against the official source, not a legal review) Official source
Singapore Policy Adopted

Singapore AI Tester Accreditation Programme

AI Tester Accreditation Programme (AI TAP)

Accredits third-party firms that test AI systems, in benchmarking and red teaming, for 12 months at a time; the first five were AIDX Tech, Knovel Engineering, PwC Singapore, Resaro and Vulcan. Accreditation does not certify that a tester is independent or that a tested system is safe.

Pending review (a factual check against the official source, not a legal review) Official source
Singapore Framework Voluntary standard

Singapore Model AI Governance Framework

Model AI Governance Framework (Second Edition) and Model AI Governance Framework for Generative AI

Singapore's Model AI Governance Framework is a voluntary, sector-agnostic guide for organisations deploying AI. The second edition (January 2020) covers four areas: internal governance structures and measures, determining the level of human involvement in AI-augmented decision-making, operations management (data, model development, monitoring), and stakeholder interaction and communication. The May 2024 Model AI Governance Framework for Generative AI extends it with nine dimensions including accountability, data, trusted development and deployment, incident reporting, testing and assurance, security, content provenance, safety and alignment research, and AI for the public good.

Adopted 21 Jan 2020 Verified (reviewed by Bhaskar Bhatt; a factual check against the official source, not a legal review) Official source
Singapore Guidance Guidance

Singapore starter kit for testing agentic AI

Starter Kit for Testing Agentic Systems for Safety and Reliability

A practical reference for organisations starting to test AI agents that reason, act on a user's behalf, use tools and access data. It sets out four steps (identify risks such as erroneous or unauthorised actions; design and run realistic tests without real-world harm; analyse results and operational logs; monitor and retest continuously), tests the agent's whole trajectory of reasoning and actions rather than only its output, and suggests metrics such as tool-use accuracy, handoff accuracy and human-intervention rates. It extends the agentic AI governance framework and builds on the LLM starter kit.

Pending review (a factual check against the official source, not a legal review) Official source
Singapore Guidance Guidance

Singapore starter kit for testing LLM applications

Starter Kit for Testing LLM-Based Applications for Safety and Reliability (v1.0)

Voluntary guidelines for testing applications built on large language models against five risks: hallucination and inaccuracy, bias in decision-making, undesirable content, data leakage and adversarial prompts. Baseline tests are available on Project Moonshot.

Pending review (a factual check against the official source, not a legal review) Official source

Frequently asked questions

What counts as an AI policy instrument here?

Anything a government or standards body has actually issued: acts, regulations, executive orders, rules, national strategies, frameworks, guidance, codes of practice, voluntary standards and open consultations. Each record states which of those it is, whether it binds anyone, and links the official source it was read from.

Are these records legally verified?

A record is marked verified only after a named reviewer has opened its official source and confirmed each field. Anything that has not been through that carries "pending review" on the record itself, and the data-health document reports the current totals. Nothing here is legal advice.

How current is the information?

Every record carries the date its source document was published and the date it was last checked. A record not re-checked within the freshness window set by the verification policy is flagged as stale rather than quietly presented as current.

Can I reuse this data?

Yes. The records are published under CC BY 4.0 with attribution, and are available as CSV, newline-delimited JSON and a read-only REST API that needs no key. Cite the record URL, the date accessed and the official source alongside it.