Resaro

Resaro's business is independent third-party testing of mission-critical AI systems — testing performance, safety, and security. Three service lines:

• Algorithm audit: independently verifying whether an AI system delivers its claimed performance
• Technical stress testing: probing system robustness under boundary conditions
• Red teaming: simulating adversarial attacks to find weaknesses

At the product level it runs Approved Intelligence, an AI assurance platform that turns testing, evaluation, validation and verification (TEVV) into reusable workflows that continuously generate deployment evidence; it has also open-sourced the testing framework ASQI Engineer.

Clients concentrate in high-stakes sectors — defence, government, critical infrastructure, healthcare (Singapore's National Healthcare Group), and financial services (Tookitaki); HTX (Home Team Science and Technology Agency), MSD and Cap Vista are on its public client list.

Ecosystem positions: founding and premier member of the AI Verify Foundation (one of only a handful globally), member of Partnership on AI, member of Germany's KI Park, and NVIDIA Inception programme member.

Resaro's bet is on the "testing gap" in AI governance: between governance frameworks (principles, guidelines, regulation) and actual deployment, there is a missing independent step that can answer "does this system actually work". Software has third-party audits, cars have TÜV — before 2023, AI had essentially no equivalent.

Its three footprints in Singapore's governance landscape:

• CSA co-authored paper: when CSA launched its Guidelines on Securing AI Systems at SICW in October 2024, the accompanying discussion paper "Securing AI: A Collective Responsibility" was co-written with Resaro — exploring how AI security differs from traditional IT security, and the new attack surfaces created by AI's adaptive nature
• Open-source contributions to AI Verify: refactored the AI Verify toolkit so tests run as independent modules, added PyTorch / TensorFlow support, with code merged into the main repository after peer review
• Healthcare AI evaluation: collaborating with PRIME-CXR (a chest X-ray triage AI) on a framework for evaluating which AI solutions deliver the highest clinical value (see the on-site speech from the launch of Singapore's Centre of AI in Medicine)

A take: testing is the "last mile" of governance. Singapore shipped the framework (AI Verify), the accreditation (the AI tester accreditation programme), and pushed the standard (ISO/IEC 42119-8) — but frameworks don't run themselves; you need commercial testing firms to turn paper requirements into executable engineering. Resaro is Temasek's capital filling exactly that gap.

Resaro is one of the entities under the "Supporting AI Diffusion" pillar of Temasek Review 2026 — the "third-party testing entity" in the report refers to it. Against Temasek's other two Singapore-based AI institutions (Temus for delivery, Aicadium for engineering), Resaro occupies the assurance/testing slot.

In the transmission levers framework:

• Lever 2 (governance): converting CSA / IMDA governance requirements into commercially procurable testing services
• Lever 6 (international + standards): dual headquarters in Singapore and Germany — one end plugged into Singapore's AI Verify ecosystem, the other into the compliance-testing demand created by the EU AI Act

A take: sovereign capital building governance infrastructure is a distinctly Singaporean pattern. The government ships frameworks and accreditation (IMDA, CSA); the state investor ships a commercial entity (Resaro); the two lines converge on the new "AI assurance" market. The arrangement gives "independent third party" a subtle wrinkle — the systems Resaro tests are often themselves Temasek-linked or government-linked.

Bottlenecks to watch: the assurance market itself is early (most enterprises haven't reached the deployment depth where third-party testing is needed), regulation-driven demand cadence (the size of the business depends on how fast AI regulation lands in each market), and scale (the team is small relative to testing demand across defence, healthcare and finance).