Governments can regulate AI while supporting innovation by tailoring obligations to the risks of particular uses, making compliance expectations clear, enabling supervised testing, coordinating standards across borders, and revising rules as evidence changes. That approach can reduce avoidable uncertainty and friction, but available evidence does not show that any regulatory model eliminates innovation costs or reliably makes AI innovation faster.
Why risk-based rules are a better fit than one rule for all AI
AI is used in settings with very different consequences when a system fails. A low-stakes consumer feature does not raise the same safety and rights concerns as an AI system used to screen job applicants or help determine access to public benefits. Regulation should therefore focus on what a system does, the context in which it is used, and the likely impact of error—not simply whether a product uses AI.
As an Amazon Associate I earn from qualifying purchases.
The European Commission describes the EU AI Act as a framework with four broad risk categories: prohibited, high-risk, limited-risk, and minimal-risk systems. Certain practices are prohibited, while high-risk systems face more extensive requirements. This illustrates how obligations can vary by risk rather than applying identically to every model or application. It is an EU approach, not a universal rulebook.
Match the obligation to the risk
- Set stronger requirements where errors could cause serious harm or affect people’s rights.
- Use lighter measures where risks are lower, rather than extending high-risk compliance duties to every AI-enabled feature.
- Make clear which entity is responsible for each obligation and how requirements interact with existing sector or product rules.
Proportionality only helps if developers can tell which category applies and what they must do. Clear guidance and predictable implementation can reduce avoidable rework; they do not mean removing safeguards.
Make compliance understandable and predictable
Developers need to know what regulators expect before they invest in building, testing, and deploying a system. Governments can publish practical guidance, explain how AI rules interact with existing laws, and provide accessible routes for questions and compliance. Predictability is especially important for smaller firms that may have limited legal and compliance capacity.
The European Commission’s account of the AI Act describes provisions intended to simplify requirements for certain smaller firms and clarify their relationship with EU product-safety laws. Those are EU-specific measures, but the broader design lesson is to reduce unnecessary duplication while keeping the underlying protections intact.
Use regulatory sandboxes for supervised testing
A regulatory sandbox is a controlled, time-limited setting in which a provider can develop or test an AI system under an agreed plan and appropriate safeguards, with regulator involvement. It can help authorities understand a technology and give participants guidance while uncertainty remains. It is not a general waiver from the law.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Rank #2
Article 57 of the EU AI Act describes sandboxes as environments for developing, training, testing, and validating innovative systems for a limited period under a plan agreed between providers and competent authorities. The provisions include guidance, risk identification and mitigation, and reporting when participation ends. Reports may help with later conformity assessment. Providers remain liable for damage, and authorities retain supervisory and corrective powers. The text allows for no administrative fines for specified regulatory infringements during participation only under stated good-faith conditions; that is not blanket immunity.
What a well-designed sandbox needs
- Clear entry criteria: Applicants should know which systems and organizations are eligible and what information they must provide.
- An agreed test plan: The plan should define the purpose, duration, safeguards, evidence to collect, and conditions for ending or changing a trial.
- Capable oversight: Regulators need the expertise and resources to give useful guidance, monitor risks, and intervene when needed.
- A real exit route: Participants need to understand what happens when a trial ends, including how findings relate to ordinary authorization or conformity assessment.
- Fair access: Application costs, selection methods, and support should not make participation practical only for large firms.
The OECD’s 2023 paper on AI regulatory sandboxes treats them as one policy tool among several and highlights evaluation, interdisciplinary cooperation, regulator expertise, interoperability, and competition effects. A sandbox does not automatically promote competition: its design can advantage incumbents if smaller firms cannot access it or meet its entry requirements.
Coordinate standards and cross-border expectations
Technical standards can turn broad legal aims into practices that organizations can test and assess. When jurisdictions coordinate standards and assessment approaches, firms may face fewer conflicting expectations across markets. Standards can support regulation, but they should not quietly replace legal accountability or public oversight.
Rank #3
NIST’s 2024 plan for global engagement on AI standards calls for international engagement and was prepared with public- and private-sector input. The OECD’s 2024 Framework for Anticipatory Governance of Emerging Technologies also identifies international cooperation in science and norm-making as part of effective governance. Together, these sources support coordination as a way to improve consistency—not as proof that one technical standard will suit every legal system or use.
Recommended Free Tools
Keep rules adaptable as technology and evidence change
AI systems, their capabilities, and the ways people use them can change faster than a fixed rule can anticipate. Governments can pair stable protections with processes for monitoring developments, assessing impacts, consulting affected groups, and reviewing implementation. That makes adaptation a continuing governance task rather than a one-time rewrite after a problem emerges.
The OECD’s 2024 framework groups anticipatory governance into five connected elements:
- Embed values in innovation.
- Use foresight and assessment.
- Engage stakeholders and society.
- Develop agile regulation.
- Cooperate internationally.
These elements work together: monitoring without public input can miss harms, while engagement without a route to revise rules may not change outcomes. Scheduled review and evidence-based adjustment can help governments respond without imposing broad new duties simply because a technology has changed.
Preserve accountability and credible enforcement
Flexible rules still need enforceable boundaries. The OECD’s 2025 Regulatory Policy Outlook says well-designed risk-based regulation can support innovation, but also notes that industry-led or co-led approaches have sometimes prioritized innovation over other regulatory objectives and left the public insufficiently protected. Industry expertise can help governments understand technical developments; it should not determine public protections without independent oversight.
Enforcement capacity matters too. Rules that are unclear, inconsistently applied, or impossible to monitor can create uncertainty for responsible firms while failing to address harmful practices. Governments should make responsibilities legible, maintain the ability to investigate and correct violations, and use evidence from implementation to improve guidance and policy.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the EU AI Act shows—and what it does not
The EU AI Act is a current example of a risk-based framework paired with innovation-support measures, including regulatory sandboxes. The European Commission’s overview, updated 3 August 2026, says the Act entered into force on 1 August 2024 and became applicable on 2 August 2026, subject to phased exceptions. It lists prohibitions and AI-literacy obligations as applying from 2 February 2025, and general-purpose AI model obligations from 2 August 2025. Following the 2026 AI Omnibus, it lists 2 December 2027 for specified high-risk use cases and 2 August 2028 for high-risk AI embedded in regulated products. These are EU dates and may be affected by later legal changes; consult the current EU legal text for a decision about a particular system.
The Commission also describes expanded sandbox access, including an EU-level sandbox, and says national authorities must provide sufficient resources and cooperate with relevant authorities. The model is intended to combine legal certainty and regulatory learning with safeguards. Its existence, however, does not establish that the Act has increased innovation or that the same design would work unchanged in another jurisdiction.
How to judge whether a regulatory approach is working
Governments should assess both protection and the conditions for responsible development. Useful questions include:
- Are obligations tied to clearly defined uses and levels of risk?
- Can developers, especially smaller firms, understand the rules and estimate compliance work?
- Do sandboxes have clear eligibility, safeguards, evaluation methods, and exit conditions?
- Do standards and conformity assessments improve consistency without displacing legal accountability?
- Do regulators have enough expertise and capacity to offer guidance and enforce requirements?
- Can firms meet requirements across borders without unnecessary duplication?
- Do monitoring and review lead to changes when evidence or technology changes?
The evidence supports these as design questions, not a proven formula for faster AI development. The OECD’s 2025 report also says that over a third of citizens in 30 countries in 2024 considered it unlikely that their national government would appropriately regulate new technologies and help businesses and citizens use them responsibly. That is a measure of public perception, not evidence that regulation increases or reduces AI innovation. The OECD’s discussion of investment associated with fintech sandboxes is likewise adjacent evidence, not a measured AI outcome.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
