Why Governance Matters for AI Startups
AI governance is no longer an abstract concern reserved for large enterprises. For AI startups in Singapore, implementing a robust AI governance framework is increasingly a commercial necessity — not because regulators require it (the framework remains voluntary), but because customers, investors, and acquirers demand it. In our practice, we have observed a clear trend: institutional investors conducting Series A and later rounds now routinely include AI governance as a due diligence workstream. Enterprise customers, particularly those in regulated sectors like financial services and healthcare, require their AI vendors to demonstrate governance compliance before signing procurement contracts. And acquirers in M&A transactions view governance maturity as a proxy for operational risk.
Singapore has positioned itself as a global leader in AI governance through a pragmatic, principles-based approach that balances innovation with accountability. The centrepiece of this approach is the Model AI Governance Framework, published by the Infocomm Media Development Authority (IMDA) in collaboration with the PDPC. Understanding and implementing this framework is one of the highest-ROI legal and compliance investments an AI startup can make.
The Model AI Governance Framework
The Model AI Governance Framework, now in its second edition, is organized around four core principles: (1) organizations using AI in decision-making should ensure that the decision-making process is explainable, transparent, and fair; (2) AI solutions should be human-centric; (3) AI governance and practices should be aligned with the organization's broader governance framework; and (4) organizations should conduct regular assessments and monitoring of their AI solutions.
For each principle, the framework provides detailed guidance on implementation. The explainability requirement, for example, does not mandate a specific technical approach — it recognizes that the appropriate level of explainability depends on the context of deployment. A credit scoring model used by a bank requires a higher degree of explainability than a content recommendation engine. The framework provides a risk-proportionate approach that allows startups to calibrate their governance investments to their actual risk profile.
The framework also addresses two issues that are particularly relevant for AI startups: data management and algorithm design. On data management, the framework recommends that organizations implement data lineage tracking, data quality assessments, and bias detection procedures. On algorithm design, it recommends that organizations maintain documentation of model design decisions, conduct regular performance monitoring, and implement human oversight mechanisms for high-impact decisions. These recommendations align closely with the requirements of the EU AI Act for high-risk AI systems, which means that Singapore startups that implement the Model AI Governance Framework will be well-positioned for EU market entry.
AI Verify in Practice
AI Verify is an AI governance testing framework and software toolkit developed by IMDA. It allows organizations to test their AI systems against internationally recognized governance principles through a structured, repeatable process. AI Verify covers eleven governance testing areas, including transparency, explainability, fairness, robustness, safety, accountability, data governance, and human agency and oversight.
The toolkit provides both process-based checks (which assess whether the organization has implemented appropriate governance processes) and technical tests (which assess the AI system's performance against specific metrics). For example, the fairness testing module allows organizations to test for demographic bias across protected attributes, using statistical metrics such as disparate impact ratio and equalized odds. The robustness testing module assesses the model's resilience to adversarial inputs and distribution shift.
From a practical standpoint, completing AI Verify testing typically requires 4-8 weeks for a mid-complexity AI system, assuming that the organization has basic documentation in place. The primary effort is in assembling the required documentation — model cards, data sheets, governance policies, and testing results — rather than in the technical testing itself. For startups that have been building with governance in mind from day one, the process is straightforward. For those that have not, the AI Verify process serves as a valuable forcing function to identify and address governance gaps.
Implementation Roadmap
For early-stage AI startups, we recommend a phased approach to governance implementation. In the first phase (months 1-3), focus on documentation: create model cards for each AI system, document your training data sources and preprocessing steps, and draft a basic AI governance policy that maps to the Model AI Governance Framework's four principles. In the second phase (months 3-6), implement technical governance: set up bias monitoring for your production models, implement logging and audit trails for AI-driven decisions, and establish a regular model review cadence. In the third phase (months 6-12), formalize and certify: complete AI Verify testing, align your governance framework with the EU AI Act requirements for your applicable risk tier, and engage external auditors to validate your governance practices. This phased approach allows startups to build governance maturity incrementally, without diverting excessive engineering resources from product development. The total investment for a seed-stage startup is typically 5-10% of one engineer's time for the first phase, scaling to 15-20% for the third phase. This is a modest investment relative to the commercial and reputational benefits of demonstrable AI governance.