Singapore’s AI Tester Accreditation Programme (AI TAP) Explained

Singapore’s Minister for Digital Development and Information, Josephine Teo, announced the programme during an international AI safety event in May 2026.

Current rollout details include:

• Launching organisation: AI Verify Foundation
• Expected rollout: second half of 2026
• Accreditation applications: expected to open around the third quarter of 2026

Why Singapore created AI TAP

The core objective of AI TAP is to increase trust in AI safety testing services.

As organisations adopt AI systems, many rely on specialised firms to probe models for weaknesses—often through red‑teaming or attempts to bypass safety safeguards. But the quality and credibility of these testing providers can vary widely.

AI TAP aims to solve this by creating a formal accreditation standard for companies offering AI testing services.

The programme is also part of Singapore’s broader effort to build a trusted AI ecosystem that balances technological innovation with safety and accountability.

Several related initiatives support this ecosystem, including:

• The AI Verify testing framework, which helps companies assess whether their AI systems follow internationally recognised governance principles.
• The Global AI Assurance Pilot and Sandbox, which connect companies deploying AI with specialised testing firms to develop practical testing methods.

Together, these programmes aim to establish shared norms for responsible AI deployment.

How AI TAP will certify AI testing companies

Companies applying for AI TAP accreditation will undergo two main categories of evaluation.

Technical capability assessment

Testing organisations must demonstrate that their evaluation methods follow recognised guidelines for AI system testing. This includes showing that their services align with guidance covering areas such as:

• Safety of large language model (LLM) applications
• Reliability and robustness testing
• Vulnerability discovery and exploit testing

Applicants must also demonstrate structured testing methodologies and appropriate datasets for evaluating AI systems.

Organisational and operational evaluation

Beyond technical expertise, AI TAP will assess the company itself. The accreditation review may examine factors such as:

• Business reputation and track record
• Operational capability and staffing expertise
• Financial stability
• Whether the services advertised match the organisation’s actual capabilities

This two‑layer review is designed to ensure that accredited testers are both technically competent and operationally trustworthy.

Types of AI risks testers are expected to examine

Although AI TAP has not published a universal checklist of tests, existing AI assurance work in Singapore highlights several risk areas commonly examined in AI systems.

These include:

• Security vulnerabilities and exploitability, including attempts to bypass safeguards or extract restricted information
• Content safety failures, such as generating harmful or inappropriate outputs
• Data leakage risks, especially in retrieval‑augmented generation (RAG) systems connected to internal knowledge bases
• Multilingual safety gaps, where guardrails that work in one language may fail in another
• Accuracy and robustness problems, including hallucinated or inconsistent responses

Testing exercises in Singapore’s AI assurance initiatives have shown that guardrails can behave differently across languages and that some system designs may expand the risk of sensitive information leakage.

Testing methods used in AI assurance

AI testing organisations typically combine automated evaluation with human analysis. Examples of methods used in Singapore‑linked AI assurance projects include:

• Adversarial prompting and red‑teaming to probe for jailbreak vulnerabilities
• Synthetic dataset generation to create edge cases and rare failure scenarios
• Accuracy benchmarking, including classification accuracy and retrieval precision
• Human‑in‑the‑loop review, where experts evaluate system responses
• LLM‑as‑judge approaches, where AI models help scale evaluation of outputs

These approaches allow testers to evaluate both the performance of AI systems and the robustness of their safeguards.

How AI TAP could shape the global AI safety ecosystem

Accrediting the testers themselves addresses a growing problem in AI governance: ensuring that AI safety evaluations are credible and consistent.

Potential impacts of the programme include:

• More confidence when organisations hire external firms to evaluate AI systems
• Clearer expectations for what constitutes competent AI safety testing
• Better comparability of testing results across industries and jurisdictions
• Stronger collaboration between AI developers and specialised safety auditors

Because AI TAP is positioned as the first accreditation scheme of its kind in Asia, it could influence how other countries structure their own AI assurance systems.

What details are still unclear

While the goals and structure of AI TAP are publicly known, several operational aspects have not yet been fully disclosed, including:

• The detailed scoring model used for accreditation
• Specific testing protocols required for certification
• Whether sector‑specific standards will apply to industries such as healthcare or finance

These details are likely to become clearer as the programme opens applications and the first wave of testing firms undergo accreditation.

The bigger picture

AI safety testing is becoming an essential part of deploying modern AI systems. Governments and companies increasingly need ways to verify whether AI models are secure, reliable, and aligned with governance principles.

Singapore’s approach focuses on certifying the testers themselves, creating a trusted layer of independent evaluators that organisations can rely on when assessing AI technologies.

If successful, programmes like AI TAP could become a key building block of the global AI governance landscape.