Singapore has adopted new lines of defence to ensure AI remains safe and trusted. Singapore's Infocomm Media Development Authority (IMDA) has announced three new initiatives to strengthen AI safety and resilience at the AI Assurance Exchange: The Agentic Edition, organised by the AI Verify Foundation (AIVF) and SGTech:
- The launch of the Starter Kit for Testing Agentic Systems for Safety and Reliability
- The launch of the AI Tester Accreditation Programme (AI TAP) by AIVF
- The formation of the Frontier AI Resilience and Security Taskforce (FAIRS)
These initiatives underscore Singapore’s commitment to advancing AI securely so that individuals and businesses can use AI responsibly and with greater confidence, IMDA said.
Starter Kit for Testing Agentic Systems for Safety and Reliability
The new kit provides a practical reference for organisations getting started with testing agentic systems. It consolidates emerging best practices and challenges of testing agentic systems gleaned from consults with government agencies and over 20 industry deployers and testers (e.g. Google, AWS, and Resaro).
It also draws from testing experiences through the Global AI Assurance Sandbox and Singapore AI Safety Institute. As testing should be contextualised to each agentic system, real-world case studies from organisations such as Cohere, CyberSierra, Earlybird AI, KBTG and Knovel Engineering are included to show how this can be done.
The Agentic Testing Starter Kit provides a structured testing approach for agentic systems:
- Identifying risks such as erroneous or unauthorised actions
- Creating realistic and robust tests that capture the spectrum of agent behaviour while minimising harmful impact
- Running tests and analysing logs
- Implementing continuous monitoring and re-testing
It also highlights what is new for agentic testing, such as testing across the agent trajectory, including both reasoning and actions, selecting agent-specific metrics such as tool use accuracy, handoff accuracy, and human-in-the-loop rates, progressively automating log analysis for key failure signals, and how to adapt testing where organisations or third-party testers have limited or black-box visibility into the system.
This guidance builds on the Model Governance Framework for Agentic AI (MGF), which provides a structured way for organisations to deploy agentic systems responsibly. Specifically, it dives deeper into the MGF’s third dimension of implementing technical controls and processes by testing agentic systems. It also extends the Starter Kit for Testing of LLM Apps for Safety and Reliability, which defined best practices for how to test large language model (LLM) apps for common risks such as hallucination, data leakage, and undesirable content.
The Agentic Testing Starter Kit will be a living document and will be updated with relevant feedback and new best practices.
AI TAP
AI Verify Foundation's first-of-its-kind accreditation programme for third-party AI testing service providers aims to build a pipeline of competent testing firms that can rigorously assess AI applications across industries, from healthcare and finance to hiring and public services. To qualify for accreditation, firms undergo technical, operational and financial evaluations to assess their testing competency, track record, delivery capability and financial sustainability.
This sets a clear bar for what good AI assurance testing service looks like. Businesses can have greater confidence in engaging accredited testing firms to identify and assess safety and reliability risks in their AI applications. This, in turn, builds greater trust in the use of AI.
The first five firms to receive accreditation are AIDX Tech, Knovel Engineering, PwC Singapore, Resaro, and Vulcan. Accreditation is for one year, which allows regular review to ensure that standards keep pace with the fast-moving AI landscape. More information can be found at https://aiverifyfoundation.sg/tester-accreditation
FAIRS
Building on efforts to address emerging risks posed by frontier AI models, IMDA has formed and convened FAIRS to strengthen the infocomm and media (ICM) sector’s resilience against AI-enabled cyberthreats. The ICM sector plays a foundational role in powering Singapore’s digital economy, with interconnected networks and services that businesses, government systems, and the public rely on every day, IMDA said.
According to IMDA, the ICM sector has a foundational role in supporting essential services. Given the potential for cyberattacks to undermine national security and economic resilience, it is important to balance the adoption of AI-powered solutions with robust cybersecurity governance against AI-driven threats. This will not only strengthen the security and resilience of our digital economy, but also provide Singapore with a significant competitive advantage, the organisation said.
Bringing together more than 20 leaders and experts from telecommunications, media, cybersecurity and AI, the taskforce will serve as a collaborative platform to strengthen the ICM sector’s defences against AI-driven cyberthreats. It will publish a report by the 1H27, with practical guidelines and best practices focusing on three priorities:
- Strengthen the response to AI-accelerated threats by improving network visibility and strengthening response capabilities to keep pace with the evolving threat landscape
- Enhance resilience against emerging risks through looking at how ICM networks and systems can be strengthened and segmented in order to protect critical services and prevent widespread disruption.
- Stay ahead of the evolving threat landscape by pooling sector-wide intelligence and expertise to identify emerging attack techniques early and translate insights into actionable guidance for cyberdefence in the ICM sector.
No comments:
Post a Comment