Your interactive prep guide for the GovTech × IMDA × Terminal 3 meetup on Responsible Data Use, PETs, and hardware-enforced trust.
These are the building blocks behind every talk tonight. Tap each card to flip it and see the deeper explanation.
A family of techniques that let you extract value from data without exposing the underlying sensitive information. Includes anonymisation, synthetic data, federated learning, differential privacy, homomorphic encryption, and TEEs. The core question: how do you use data you are not allowed to see?
Stripping or obscuring personally identifiable information (PII) from datasets so individuals cannot be re-identified. Techniques include generalisation (reducing precision), suppression (removing fields), and k-anonymity (ensuring each record blends into a group of k). Risk: insufficient anonymisation can lead to re-identification attacks.
AI models generate artificial datasets that preserve the statistical properties of the original data without containing any real personal records. Use case: train ML models on sensitive data (e.g. healthcare) without exposing patient PII. Key challenge: utility-privacy trade-off — more realistic synthetic data risks leaking patterns from the original.
Instead of sending data to a central server, the model is sent to where the data lives. Each node trains locally and shares only model updates (gradients), not raw data. The central server aggregates updates to improve the global model. Used by Google (Gboard), healthcare networks. Adds differential privacy noise to updates to prevent reverse-engineering of training data.
A hardware-isolated enclave within a processor (e.g. Intel SGX, AMD SEV, ARM TrustZone) where code and data are protected from the host OS, hypervisor, and other processes. Even the server owner cannot see inside. Terminal 3's thesis: TEEs provide hard-gating — the AI agent physically cannot access data outside its enclave, no matter how clever its reasoning.
A cryptographic method where you prove you know something without revealing what you know. Example: prove you are over 18 without showing your birthdate. In AI: prove an agent ran the correct model on the correct data, without exposing the model or data. Combined with TEEs, creates a verifiable trust chain — the agent provably did what it claims, on data it provably never exfiltrated.
GovTech's Responsible Data Use framework assesses data risk across the AI lifecycle. Drag the slider to see which PETs are appropriate at each risk level.
Data Sensitivity Level
Terminal 3's core argument: AI agents need hard-gating (hardware enforcement), not just software guard-railing. Can you tell which is which?
Match the real-world scenario to the right privacy-enhancing technique. Test your understanding before the talks.
Four talks, one Q&A. Here's who's speaking and what to expect from each session.
Prepared questions make you memorable. Pick 2-3 from below that match your interests. Each is targeted at a specific speaker.
Tap to check off as you go. Show up prepared, not just present.