Singapore Global AI Assurance Sandbox

Applies To

Overview

The Global AI Assurance Sandbox is an international initiative of IMDA and the AI Verify Foundation, a non-profit subsidiary of IMDA. It was launched on 7 July 2025, building on the Global AI Assurance Pilot announced in February 2025. It matches builders and deployers of generative AI applications, including agentic AI, with specialist technical testers. Applications must be live or intended to go live. The focus is technical testing of the application, not process governance or the underlying foundation model. It is not a regulatory sandbox and offers no waivers or relaxed rules. The commercial testing partner runs the tests in the participants' own environments, and the two sides agree terms commercially. Test results stay confidential unless the participant chooses to share them, and participants publish a case study report. The testing baseline is IMDA's Starter Kit for Safety Testing of LLM-based Applications. It covers risks such as hallucination, bias, undesirable content, data leakage, and adversarial prompts. Other Singapore regulators can take part as observers to inform their own testing expectations.

Key Requirements

• •Open to builders and deployers of generative and agentic AI applications that are live or about to go live.
• •Participants are matched with specialist technical testing firms and agree terms commercially.
• •Testing covers the application, not the underlying foundation model or process governance.
• •IMDA's Starter Kit for Safety Testing of LLM-based Applications is the baseline guidance.
• •Test results stay confidential to the participants unless they choose to share them.
• •Participants publish a case study report, and the sandbox grants no regulatory relief.

What Your Organization Must Do

• →Treat sandbox participation as voluntary testing support, not approval or any relaxation of Singapore law.
• →Review IMDA's Starter Kit for Safety Testing of LLM-based Applications as a baseline for your own testing.
• →Check whether your generative or agentic AI application is live or close to launch before applying.
• →Negotiate scope, timing and cost directly with the matched testing firm, since terms are commercial.
• →Plan for publishing a case study report, and confirm internally what detail you will disclose.
• →Brief legal and risk teams that test results stay confidential unless your organisation chooses to share them.

Playbook Guidance

Step-by-step implementation guidance for compliance teams.

Frequently Asked Questions

Does participating in the Singapore Global AI Assurance Sandbox constitute regulatory approval of an AI system?

No, the sandbox is a testing and matching initiative, not a regulatory sandbox, and it grants no approval or relief.

Who can apply to the Global AI Assurance Sandbox and is it open to companies outside Singapore?

Builders and deployers of generative and agentic AI applications, in Singapore or elsewhere, apply through an expression of interest. Specialist testing firms can also join as testing partners.

What reporting obligations apply to sandbox participants?

Participants publish a case study report. Test results stay confidential between the participant and the tester, and IMDA and the AI Verify Foundation do not see them unless the participant agrees.

How long does sandbox participation last?

There is no fixed sandbox period. The scope and timing of each engagement are agreed commercially between the participant and its testing partner.

How does the sandbox relate to Singapore's Model AI Governance Framework?

Testing uses IMDA's Starter Kit for Safety Testing of LLM-based Applications as its baseline. Like the Model AI Governance Framework, the Starter Kit is voluntary guidance.

Are any rules relaxed for applications in the sandbox?

No, the sandbox never relaxes any rules, so nothing changes at deployment. Existing Singapore law applies to the application throughout.