Good morning, and welcome to day two of the Singapore Data Festival. Thank you for joining us today.
This year’s festival comes at an especially important moment. With AI, data has gained greater prominence. AI is rapidly transforming how organisations use data, accelerating innovation on the one hand, but also the scale of privacy and data governance risks on the other. Organisations are, for example, deploying increasingly capable AI systems and AI agents that can access internal systems, customer data and business processes. So that raises productivity significantly, but without appropriate safeguards, these systems can create new attack surfaces and expose sensitive information at an unprecedented scale.
As AI systems become more autonomous and increasingly connected to sensitive data, internal systems and critical business processes, the risks are heightened. In this environment, the role of Privacy Enhancing Technologies, or PETs, will become increasingly important. They are not simply a niche set of technologies but are becoming essential infrastructure for trusted and responsible data use. The use of PETs addresses the growing challenges in two ways.
First, they allow organisations to collaborate and derive insights without moving or exposing potentially sensitive original data. Two years ago, Apple introduced “Private Cloud Compute” for Apple Intelligence, allowing more computationally intensive AI tasks — such as summarisation and writing assistance to be securely conducted in the cloud environment, while ensuring that even Apple cannot access a user’s personal data during processing.
Second, where sensitive data must be shared, the use of a PET ensures that the underlying data remains protected while in transit, in storage or in use. Companies such as IBM have been exploring the use of Fully Homomorphic Encryption, or FHE, to enable AI models to perform inferencing directly on encrypted data. This allows sensitive information — such as healthcare or financial data — to remain protected even while being analysed by AI systems.
While the technology is still evolving and improving, momentum has been growing across the industry. According to the OECD, research activity in PETs has accelerated rapidly, with the number of PET-related publications increasing close to eighteen-fold over the past decade.
Techniques such as Federated Learning and Synthetic Data have especially been gaining traction because they enable organisations to develop and use AI systems while reducing the exposure of sensitive data. On the enterprise side, many technology companies have already begun deploying PETs at scale. For example, through its NeMo platform, NVIDIA enables the creation of synthetic data to help train Agentic AI systems while reducing reliance on sensitive real-world datasets. Nevertheless, organisations still face challenges adopting PETs due to its technical complexity, implementation costs and uncertainty on how the use of PETS would comply with regulations.
To address these challenges, we launched the PET Regulatory Sandbox in 2022 — as a safe environment for organisations to explore how PETs can address business needs while meeting regulatory requirements. Since its launch, we have worked with ten companies on their PET implementation, providing technical and regulatory guidance. This year, we continue to see strong interest in AI-related applications, as well as use cases in sectors such as finance and healthcare. Allow me to share one specific example in each of these two sectors.
In healthcare, medical images typically contain sensitive information about a patient, and accordingly, its use outside of a secure hospital environment is restricted. For this use case, the Singapore General Hospital (SGH) collaborated with A*STAR to explore the use of a cloud based Trusted Execution Environments as an additional layer of protection to conduct secure inferencing of an AI model using medical images. The use case is interesting, because it pushes the boundaries on leveraging PETs to make available sensitive data to be used for AI in a responsible and trusted manner.
In finance, Ant International is collaborating with us on its second use case, enabling the use of Multiparty Computing (MPC) to evaluate business risk controls for payments processing. The use of MPC ensures commercially sensitive datapoints do not need to be collected and shared between collaborating parties, but yet, can still be effectively analysed and evaluated to determine a risk control result.
To further support industry adoption, we will be enhancing the PET Sandbox programme. PET Sandbox 2.0 will focus on three areas.
First, to simplify understanding and lower technical complexity for prospective end users, we will be collaborating with PETs solution providers to co-create “PETs Demos”. These demos will be designed to be use-case centric, lightweight, easy to use, to ensure users can quickly appreciate how the PET works and learn how it could be contextualised to their own business or use case. Several partners whom we have collaborated with on these demos will share more in today’s sessions and at the booths outside. We encourage you to check them out at your own time.
Second, many organisations remain uncertain about how PETs fit within existing regulatory frameworks. To provide greater clarity, we will work more closely with sector leads across the Government to develop sector-specific guidance on PETs use.
Finally, to reduce implementation costs, we will continue offering grants to offset the costs involved in PETs Proof of Concept.
To ensure secure and effective PETs implementation, IMDA is also strengthening support for responsible data use in AI through PETs technical guides. We will be launching a Federated Learning Technical Guide that was jointly developed with GovTech Singapore, and an updated Synthetic Data Generation Guide. The two guides will complement the PETs adoption guide, providing practitioners with deeper guidance on implementation considerations and common pitfalls.
Later today, we will also be organising a Federated Learning workshop, bringing together CDOs, CAIOs to share and understand real world implementation experiences. The update to our Synthetic Data Generation Guide will reflect the latest methods and emerging use cases from organisations such as NVIDIA and GovTech Singapore. Synthetic data has traditionally been used for structured, tabular datasets, but generating realistic unstructured data — especially medical imaging data - remains more challenging. So, to advance innovation in this space, we have launched an innovation call with SGH earlier in the month, challenging solution providers to generate more synthetic medical images to make available more data to train and improve the accuracy of AI models for anomaly detection. We encourage those with applicable solutions to apply.
However, advancing PETs domestically is only part of the picture. As you may have heard in Minister Josephine Teo’s opening address yesterday, AI development and data collaboration have become increasingly cross border and international collaboration will become equally important in ensuring trusted and secure data sharing. To support this, IMDA has commissioned an ASEAN PETs Report in collaboration with our regional partners. The report will examine how PETs can support trusted cross-border data sharing and encourage greater collaboration and PETs use across the region. We also welcome the OECD’s Global PETs Repository initiative, aimed at improving awareness, understanding and adoption of PETs globally. Collectively, such efforts will help build a stronger and more connected global PETs ecosystem
AI will bring significant opportunities for economies and societies. But as AI systems become more capable and connected, PETs will become essential infrastructure for data protection whilst ensuring trusted innovation. But just to be clear, PETs, on their own, are not a silver bullet. They do not replace the need for sound governance, cyber security, access controls and human oversight. Rather, they should be seen as one important layer within a broader trusted data ecosystem.
The message at last year’s PETs Summit centres around how PETs are no longer a compliance tool, but a competitive differentiator. This continues to be true. But with the pace at which AI capabilities — and corresponding risks — continue to evolve, there is now a greater sense of urgency. Organisations that adopt PETs will not only be better protected, but also better positioned to innovate and collaborate with confidence.
With that, thank you for your attention and I wish you a fruitful session ahead.