New Singapore scheme to certify firms that test and jailbreak AI systems

SINGAPORE – A new scheme to accredit companies that test artificial intelligence systems, said to be the first in Asia, will be launched in Singapore by the third quarter of 2026.

The AI Tester Accreditation Programme (AI TAP), by the not-for-profit AI Verify Foundation, will benefit companies looking to hire external testers to jailbreak their AI products and uncover weaknesses before deployment.

The programme was announced by Minister for Digital Development and Information Josephine Teo on May 18 at the International Scientific Exchange on AI Safety 2026.

AI Verify Foundation, a subsidiary of the Infocomm Media Development Authority (IMDA), said: “The programme aims to build a pipeline of competent testing firms that can rigorously assess AI systems across industries, from healthcare and finance, to hiring and public services.”

The foundation pointed out that businesses need to verify that their AI systems are trustworthy, reliable and operate as needed.

But because there is no clear definition of a qualified AI testing firm, it is difficult to identify capable companies.

“This gap is particularly obvious for technical testing services, which require specialised expertise and remain in short supply. This creates an opportunity for Singapore to contribute to setting the standards on good-quality testing,” the foundation said.

Speaking at the event, Mrs Teo said that as Singapore seeks to become an AI hub, it will focus on creating an environment of trust, where technology is deployed responsibly, risks are well understood, and effective protections thoughtfully implemented.

“This matters especially because AI is moving beyond chatbots and search tools. Increasingly, even in Singapore, we see AI systems being used in areas that affect people’s lives directly: healthcare, finance, transport, public services, education, cybersecurity and critical infrastructure,” she said, adding that a trusted AI ecosystem may ultimately become more attractive than a purely fast-moving one.

She noted that Singapore has introduced initiatives such as AI Verify to encourage responsible AI testing and transparency, and the new programme to accredit third-party AI testers will give deployers added confidence.

“The scheme to accredit third-party AI testers, who are not the Government, is intended to set standards and to uphold a certain level of confidence in the testers themselves,” said Mrs Teo.

Companies hire AI testers to simulate how bad actors could try to trick AI systems into breaking their own safety rules before the products are released publicly.

This can include attempts to make chatbots generate malware code, reveal confidential information or give dangerous advice that the systems were designed to block.

Common testing methods include using carefully crafted prompts to make AI systems ignore their original instructions and built-in safety safeguards, hiding such malicious prompts in uploaded files or webpages, and pretending to be someone with higher administrative rights.

Under the new scheme, AI testing companies will be assessed to ensure technical competency based on guidelines such as the IMDA Starter Kit for Testing LLM-Based Applications for Safety and Reliability.

Published in January, the guidelines set out the five key risks in large language models, the types of tests that should be conducted for those risks, and how to analyse and interpret the results.

The testers will also be assessed to ensure they have good standing and are financially sustainable and operationally ready.

No application or accreditation fees will be charged under the scheme.

Testing companies such as Advai, AIDX, Ernst & Young, Knovel Engineering, PwC, Resaro and Vulcan have expressed early interest in the programme.

Vulcan co-founder Alex Leung said the accreditation will help create clearer standards around what good testing looks like.

He said that right now, many testers may simply take open-source benchmark data sets or generic jailbreak prompts and run them against a client’s AI system.

“That can be useful as a starting point, but it is not enough. Proper AI testing has to be customised to the specific application, its use cases, its connected tools, its data flows, and the real-world threat scenarios it faces,” Mr Leung said.

Knovel Engineering chief executive Seah Hee Chuan said many organisations today claim to have rigorous AI testing methodologies, but the depth and breadth of each organisation vary.

“Accreditation helps in several ways – establishing a baseline competency for accredited testers, ensuring governance, and standardising methodologies,” he said.