Prepared with AI assistance from the linked primary sources. Company and government claims are attributed to their issuers. How we use AI.

A useful AI policy comparison starts with a product, not a league table. A company deciding where to offer an employee assistant needs different information from a bank introducing an automated customer service. The same country can have several relevant instruments, each addressing a different part of the operation.

1. Define the activity before comparing rules

China’s Interim Measures illustrate the importance of audience: their scope concerns generative services offered to the domestic public, with an exclusion for specified activities that do not offer such public services. A comparison that records only that China has generative-AI measures misses this boundary. China’s Interim Measures for Generative AI Services, 2023.

For a hypothetical deployment, identify what the system produces, who can access it and whether it can take action. Then record which organisation supplies the model, operates the application and handles the data. These are the facts a policy assessment must relate to the relevant text.

2. Match the instrument to its audience

Hong Kong’s monetary authority issued its August 2024 generative-AI consumer-protection circular to authorised institutions. It addresses customer-facing applications and sets expectations concerning governance, fairness, disclosure and protection of customer information. It should not be presented as a general statute covering every AI developer. HKMA circular on consumer-facing generative AI, 19 August 2024.

This distinction prevents a common comparison error: placing a banking supervisor’s circular beside a national law and treating them as equivalent. The authority, covered organisations and legal mechanism all matter. So does the relationship with other documents that the instrument references.

3. Separate shared practice from legal reach

The ASEAN guide offers voluntary regional guidance while preserving national legal obligations. Its existence supports comparison of governance practices; it does not establish identical national requirements or automatic recognition of one country’s compliance decision in another. ASEAN Guide on AI Governance and Ethics, 2024.

Organisations can benefit from using a common internal vocabulary, but should retain a clear distinction between a useful control and the reason it is required. A practice might be adopted voluntarily, incorporated into a customer contract or connected to a specific supervisory expectation. Those routes have different consequences.

4. Look for implementation evidence

GovTech Singapore’s evaluation guidance moves from broad principles to a defined use case, representative tests, measures of success and analysis of failures. It covers functional performance alongside safety, robustness, fairness and privacy. GovTech Singapore guidance on evaluating AI systems.

That provides a useful comparison dimension: how does an organisation demonstrate that its policy works? A statement about human oversight means little unless the reviewer receives relevant information, has time to intervene and can change the outcome. The evidence should fit the action the system is permitted to take.

What a decision-ready comparison contains

A practical briefing should identify the instrument, its status, the activity it reaches, the responsible authority and the next implementation milestone. It should also explain what remains uncertain. A consultation response, an effective rule and an enforcement decision provide different evidence.

The takeaway is to compare like with like. Watch for changes that affect the actual service: a revised definition, a new disclosure duty, a sector circular or an updated technical requirement. The result will be more useful than counting documents or labelling an entire market friendly or restrictive.

Sources

See something we should correct? Contact the newsroom.