Global AI Assurance Sandbox
Singapore · Framework · IMDA and the AI Verify Foundation
Our summary
An international initiative that brings together AI application developers and specialist testing firms to conduct real-world testing and assurance of generative AI systems. Participants collaborate to identify testing requirements, conduct technical assessments, and produce case studies that help inform standards development. The programme aims to reduce adoption barriers, foster testing market growth, and generate practical guidance for AI assurance.
Written by the Index in our own words. It is not legal advice; the official text is the authority.
Inside this policy
Every section of this policy in order. Our notes summarise each one in plain language and do not quote the official text, which is the authority.
Drafted by the Index An automated summary of the official text, not yet reviewed by a person.Reviewed Checked by a person against the official text.
16 sections
Jump to a chapter
Connect AI Deployers and Testers to Enable Real World AI Testing and Assurance
IMDA and the AI Verify Foundation jointly operate this international sandbox where organizations that build or deploy generative AI applications can engage qualified testing specialists to assess their systems. The program continues beyond its initial pilot phase, accepting new participants from both the deployment and technical testing communities.
Objectives of Sandbox
1 section
One objective of the Sandbox is to reduce testing-related barriers that currently hinder the adoption of generative AI.
Who Should Join the Sandbox and How Can It Help You?
4 sections
Why join?
AI deployers launching or deploying generative or agentic AI applications can benefit from obtaining testing guidance, identifying testing partners, showcasing their use cases, and accessing funding for specialist testing expertise.
What do you need?
Applicant AI deployers must have generative applications based on large language models, multi-modal models, or agentic systems, excluding those relying on older video, image, or voice technologies. Applications must already operate in live environments or be ready to do so, not remain in experimental stages.
Why join?
AI testing service providers can participate to connect with potential customers, validate their testing approaches on real use cases, demonstrate their capabilities, and participate in developing emerging standards for generative AI testing.
What do you need?
Applicant testing firms must provide assessment services or tools for AI systems and possess demonstrated capabilities in establishing and conducting various testing approaches such as benchmarking and adversarial testing. These firms focus exclusively on technical evaluation of deployed applications rather than governance processes or the underlying foundation models themselves.
Testing Focus Areas, Identify What Matters Most to Test
The IMDA Starter Kit helps participants identify testing priorities including hallucination, bias, undesirable content, data leakage, and adversarial robustness, plus use case-specific considerations relating to safety, financial impacts, human oversight, and regulatory compliance.
Sandbox Output
Participants will produce a case study report and receive public recognition through website features and event participation.
The Sandbox
1 section
The Sandbox process consists of expressing interest, assessment of qualifications, and then proceeding through introduction, technical testing, and case study sharing phases.
Introduction
Technical Testing
Case Study Sharing
Resources
What Do Participants Say
Controls
What this policy requires, forbids, grants or penalises, one control at a time and in our own words. Each control links to the official text, which is the authority. This is not legal advice.
Drafted by the Index Drafted automatically from the official text, not yet reviewed by a person.Reviewed Checked by a person against the official text.
Themes covered 6 of 35
Choose a theme to see only its controls. About the themes
Risk and scope 1 of 4
- Risk classification, not covered
- Prohibited practices, not covered
- Impact assessment, not covered
Data and privacy 0 of 4
- Data governance and quality, not covered
- Data subject rights and consent, not covered
- Cross-border data transfer, not covered
- Copyright and training data, not covered
Transparency and records 2 of 4
- Labelling of synthetic content, not covered
- Record keeping and logging, not covered
Technical safeguards 1 of 4
- Human oversight, not covered
- Accuracy, robustness and cybersecurity, not covered
- Bias and non-discrimination, not covered
Assurance and lifecycle 0 of 4
- Conformity assessment and certification, not covered
- Registration and notification, not covered
- Incident and breach reporting, not covered
- Post-market monitoring and corrective action, not covered
Rights and protected groups 0 of 4
- Automated decision-making rights, not covered
- Children and vulnerable groups, not covered
- Biometric identification and surveillance, not covered
- Whistleblowing, complaints and redress, not covered
Roles and organisation 2 of 6
- Supply chain and provider-deployer duties, not covered
- General purpose and foundation models, not covered
- Accountability and governance roles, not covered
- Public sector use and procurement, not covered
Enforcement and sectors 0 of 5
- Enforcement, penalties and regulators, not covered
- Sector rules: health, not covered
- Sector rules: finance and insurance, not covered
- Sector rules: employment, not covered
- Sector rules: education, not covered
The full list of controls could not be loaded, so only the first 18 are shown.
Connect AI Deployers and Testers to Enable Real World AI Testing and Assurance 1 control
- ObligationApplies to: Public authorityDrafted by the Index
IMDA and the AI Verify Foundation operate an international sandbox where organizations building or deploying generative AI applications can engage technical testing specialists to assess their systems.
Reduce testing-related barriers to GenAI adoption through 3 controls
- ObligationApplies to: Public authorityDrafted by the Index
Public authorities should provide guidance to help reduce obstacles to testing AI systems in sandbox settings.
- ObligationApplies to: Public authorityDrafted by the Index
Public authorities should give testing companies opportunities to help shape technical standards for generative AI tools.
- ObligationApplies to: Public authorityDrafted by the Index
Public authorities should help build a sustainable market for AI testing and assurance services.
Why join? 2 controls
- RightApplies to: DeployerDrafted by the Index
AI deployers launching or deploying generative or agentic AI applications may obtain guidance on what and how to test, introductions to testing partners, opportunities to showcase use cases, and access to funding for specialist testing.
- Applies to: ProviderDrafted by the Index
Testing service providers may join the sandbox to connect with customers, validate their methodologies, showcase capabilities, and help develop standards for generative AI testing.
What do you need? 5 controls
- ObligationApplies to: DeployerDrafted by the Index
Applications admitted to the sandbox must use large language models, multi-modal models or agentic systems, but not rely on pre-LLM video, image or voice technologies.
- ObligationApplies to: DeployerDrafted by the Index
Generative AI applications in the sandbox must operate in live production environments or be ready to do so, rather than remain purely experimental.
- ObligationApplies to: ProviderDrafted by the Index
Testing firms must deliver AI testing products or services to participate in the sandbox.
- ObligationApplies to: ProviderDrafted by the Index
Testing firms must show proven ability to design and implement AI testing methods such as benchmarking and adversarial testing.
- ObligationApplies to: ProviderDrafted by the Index
Testing firms must focus on technical evaluation of deployed systems rather than governance or the underlying foundation models.
Testing Focus Areas, Identify What Matters Most to Test 2 controls
- ObligationApplies to: ProviderDrafted by the Index
Sandbox participants should use the IMDA Starter Kit to identify testing priorities across hallucination, bias, undesirable content, data leakage, and adversarial robustness.
- ObligationApplies to: ProviderDrafted by the Index
Sandbox participants should explore use case-specific testing considerations including safety, financial impacts, human oversight, and regulatory compliance risks.
Sandbox Output 2 controls
- ObligationApplies to: ProviderDrafted by the Index
Sandbox participants must produce a case study report documenting their testing work.
- RightApplies to: ProviderDrafted by the Index
Sandbox participants will receive public recognition on the organiser's website and at relevant events.
Introduction 1 control
- Applies to: AnyoneDrafted by the Index
AI deployers and testers should be matched with each other according to their objectives, scope and preferences, and partners should be selected through commercial discussion.
Technical Testing 1 control
- ObligationApplies to: AnyoneDrafted by the Index
Participants should complete technical testing and produce a case study report within three months.
Case Study Sharing 1 control
- ObligationApplies to: AnyoneDrafted by the Index
Case study reports should be submitted to IMDA and AIVF and then published.
Change history
No changes recorded since we began monitoring this policy.