Dominic Jainy stands at the forefront of the technological revolution in Southeast Asia, bringing a wealth of experience in artificial intelligence, machine learning, and the strategic implementation of blockchain. As an IT professional who has navigated the complexities of enterprise-scale deployments, Jainy offers a unique perspective on how cutting-edge models are moving from experimental labs into the heart of national infrastructure. His insights are particularly timely as Singapore establishes itself as a global epicenter for AI innovation, serving as a primary base for some of the world’s most advanced frontier model developers.
This conversation delves into the rapid integration of Claude across Singapore’s public and private sectors, exploring the technical nuances of high-scale adoption and the rigorous safety frameworks required for national security. We explore the massive efficiency gains seen by regional giants, the localized architectural choices for regulated banking environments, and the critical role of multilingual safety testing in a diverse society.
Singapore currently ranks second globally in Claude.ai usage relative to population size. What specific cultural or economic factors drive this high adoption rate, and how do you tailor your regional strategy to meet the demands of a market that outpaces international benchmarks by nearly six times?
The high adoption rate in Singapore is a direct reflection of the nation’s relentless pursuit of efficiency and its “Smart Nation” ethos that permeates both the government and the workforce. When you see adoption rates nearly six times the international benchmark, it’s a clear signal that the local professional community views AI not as a novelty, but as an essential utility for survival in a high-cost economy. To meet this demand, we’ve shifted our strategy toward providing extreme reliability and security, because users here are putting the technology to work on their most difficult, high-stakes problems. There is a palpable energy in the local developer community—as evidenced by the 700 builders who gathered at our recent event—which forces us to maintain a team on the ground to provide immediate, localized support.
GovTech Singapore has deployed Claude Code among more than 1,500 technical personnel to modernize internal systems. What are the primary technical hurdles when integrating AI into government infrastructure, and what step-by-step protocols ensure these tools accelerate software delivery without compromising public sector security?
The primary hurdle is the sheer complexity of modernizing legacy government systems that were built long before generative AI was a reality. With 1,500 technical personnel at GovTech now using Claude Code, we have to ensure that these tools are integrated into a secure software development lifecycle that includes rigorous automated checks and human oversight. Our protocols involve a layered approach: first, utilizing the Generative AI Evaluation Sandbox to test models in a controlled environment, followed by formal technical exchanges with domestic security organizations like the Cyber Security Agency. This ensures that as we accelerate delivery schedules, we are simultaneously hardening the infrastructure against vulnerabilities. It’s about creating a “safety-first” culture where the speed of development is balanced by a robust, multi-agency audit trail.
Grab’s internal sales assistant, BriX, reportedly saves representatives 40,000 hours per quarter. Beyond time-saving metrics, how does this level of automation shift the day-to-day responsibilities of human staff, and what specific advice do you have for other regional enterprises looking to replicate this ROI?
Saving 40,000 hours every three months fundamentally changes the “emotional labor” of a sales representative by removing the exhausting drudgery of manual data searching and administrative overhead. This shift allows employees to focus on high-value cognitive tasks, such as building deeper relationships with merchants or solving complex logistical challenges that require human empathy and intuition. For other regional enterprises, my advice is to identify a specific, data-heavy “friction point” within your operations and pilot a specialized tool like BriX rather than a generic chatbot. Success comes from making the AI feel like a seamless extension of the team’s workflow, allowing the staff to feel empowered rather than replaced by the technology.
Regulated institutions like DBS utilize in-region inference via Amazon Bedrock to maintain strict data compliance. How does this architecture balance the need for high-performance computation with localized legal requirements, and what trade-offs must engineering teams consider when choosing between on-shore and cloud-based deployments?
For a massive institution like DBS, Southeast Asia’s largest banking group, the architecture must satisfy the dual gods of performance and sovereignty. In-region inference via Amazon Bedrock is the solution that allows them to tap into massive computational power while ensuring that sensitive financial data never leaves Singapore’s borders. Engineering teams have to weigh the trade-off between the slightly higher latency of certain localized cloud setups against the absolute necessity of meeting strict MAS (Monetary Authority of Singapore) regulations. It’s a delicate balance, but by keeping inference “on-shore,” banks can achieve the reliability their customers demand while maintaining a fortress-like security posture that is required for mission-critical commercial operations.
Multilingual red-teaming assessments have already been conducted in English, Mandarin, Malay, and Tamil. What unique safety risks or linguistic nuances were uncovered during these trials, and how do these findings influence the technical exchanges between AI developers and domestic security organizations like the Cyber Security Agency?
Testing in Singapore’s four official languages—English, Mandarin, Malay, and Tamil—uncovered that safety risks aren’t always universal; they can be buried in cultural idioms or specific linguistic structures that a mono-lingual model might miss. These red-teaming trials revealed how “jailbreaking” attempts or harmful content can be masked through code-switching or regional slang, which is common in a multicultural hub. These findings are shared directly with the Cyber Security Agency of Singapore to refine our defensive layers and ensure the model understands the local context. This ongoing technical exchange creates a feedback loop where the AI becomes more “culturally aware,” making it far more resilient against adversarial attacks that target localized nuances.
Small and medium-sized enterprises (SMEs) are now a focal point for operational adoption studies. What are the most common barriers for smaller firms trying to implement frontier models, and what specific resources or frameworks are being established to ensure these businesses aren’t left behind by larger industry operators?
The most common barriers for SMEs are often a lack of internal technical expertise and the perceived high cost of entry for integrating frontier models into their existing workflows. To bridge this gap, we are participating in operational adoption studies that provide a framework for smaller firms to test AI tools without a massive upfront investment. By working with the Infocomm Media Development Authority and the AI Verify Foundation, we are helping to democratize access to the same safety and evaluation tools used by giants like Sea or Grab. Our goal is to ensure that a 20-person engineering firm in Singapore has the same opportunity to leverage Claude Code as a multinational corporation, preventing a digital divide in the local economy.
Your expansion includes deep engagement with third-party evaluators like METR and platforms like Hugging Face. How do these external safety audits function in practice, and what specific metrics are used to determine if a model is “reliable” enough for deployment in mission-critical banking or infrastructure sectors?
External audits by groups like METR function as an independent “stress test,” where they push our frontier models to their limits using adversarial scenarios that we might not have conceived of internally. They look at metrics such as “autonomy risk,” “hallucination frequency,” and the model’s ability to resist manipulation in high-pressure environments. When a model is being considered for a mission-critical role in banking or public infrastructure, it must exceed specific thresholds for reliability and predictable behavior. This collaboration with platforms like Hugging Face ensures that we are part of a global, transparent conversation about AI safety, providing the “trust layer” that Singaporean organizations demand before they integrate AI into the core of their operations.
What is your forecast for AI deployments in Singapore?
I predict that by the end of 2026, we will see the total integration of AI into the “physical” infrastructure of the city, where Claude and similar models aren’t just answering emails but optimizing real-time logistics and public utility grids. The partnership between the public sector and private AI firms will evolve into a unified “national operating system,” where data flows seamlessly between agencies to provide proactive services to citizens. We are moving toward a future where “AI-native” will be the default state for every new business started in Singapore, further cementing the country’s position as the most AI-ready nation on the planet. The synergy between human policy and machine intelligence will become so tight that the distinction between “digital services” and “AI services” will effectively disappear.