The Two-Speed Approach to AI Rules
The central idea gaining traction in government and corporate boardrooms is a two-phase approach to artificial intelligence governance. This strategy involves a deliberate, slow-moving initial phase focused on experimentation and learning, followed by
a rapid scaling phase once safety and effectiveness are better understood. The logic is that AI technology is evolving too quickly for traditional, slow-moving legislative processes to keep up. By the time a comprehensive law is passed, the technology it was designed to regulate may already be obsolete. Instead of a single, all-encompassing law, this model proposes a more adaptive framework: learn in a controlled environment, and then apply those lessons to enable broad, but safe, deployment. This 'crawl, walk, run' method is seen as a way to balance the immense pressure for innovation with the growing concerns over AI's potential risks, from biased decision-making to significant security vulnerabilities.
Learning Slowly Through Regulatory Sandboxes
The 'learn slowly' phase is primarily being implemented through a tool called a 'regulatory sandbox'. A sandbox is a controlled legal environment where companies can test new AI technologies under the supervision of regulators, often with temporary relief from certain existing rules. The European Union's AI Act mandates the creation of these sandboxes, and similar initiatives are being proposed in the United States and piloted in countries like the UK and Singapore. The goal is to close the information gap between innovators and policymakers. Developers get a chance to test their products in real-world conditions with reduced legal uncertainty, which can accelerate product iteration and lower costs. In return, regulators gain invaluable insight into how these complex systems behave, allowing them to gather evidence to inform future, more effective, and evidence-based rules. This collaborative process helps ensure that AI systems are tested for safety, fairness, and transparency before they are deployed to the wider public.
The Blueprint for Scaling Quickly
Once an AI application has been tested and its risks are better understood, the governance model shifts to enabling rapid and responsible scaling. This phase relies on establishing clear standards and best practices learned from the sandbox phase. Frameworks like the NIST AI Risk Management Framework in the US and the international ISO 42001 standard provide voluntary but influential guidelines for organizations to build their governance programs around. These frameworks help standardize documentation, risk management, and oversight. For companies, adhering to these standards can streamline compliance, build investor confidence, and provide a competitive advantage. The idea is that by proving their systems are safe and aligned with established best practices in a controlled setting, companies can then move much faster to deploy their innovations across markets, with a lower risk of encountering regulatory roadblocks or causing unintended harm.
A Fragile Consensus
While this phased approach is gaining support, it is not without its critics and challenges. A major concern is that innovation continues to outpace even these more adaptive regulatory efforts. Some argue that a 'slow start' still allows technology to race ahead while governance plays catch-up. There's also the risk that regulatory sandboxes could be captured by large, well-resourced companies, giving them an unfair advantage and potentially stifling competition from smaller startups. Furthermore, many current AI governance strategies within organizations are criticized for being performative—creating a false sense of security with policies that aren't backed by technical enforcement or genuine executive accountability. The gap between having a written policy and having effective, real-time controls remains a significant problem, especially as AI systems become more autonomous. The success of the 'learn slowly, scale quickly' model depends entirely on rigorous oversight and a genuine commitment to translating lessons into binding rules, not just creating loopholes for favored players.
