Commentary. The central question in artificial-intelligence policy is no longer whether governments will regulate AI. They are already doing so. The more consequential question is whether regulation will become a dependable public safeguard—or an exhausting patchwork that leaves citizens confused, companies guessing and regulators chasing yesterday’s systems.
Our view is clear: governments should resist both reflexive deregulation and the temptation to legislate every new feature separately. The better course is enforceable, risk-based rules, coordinated across borders and tested against real-world harms.
The urgency is evident. The European Union’s AI Act entered a major enforcement phase on Aug. 2, 2026, giving the AI Office and national market-surveillance authorities powers covering general-purpose models and other AI systems. The European Commission has also opened a consultation on procedural rules for investigating and penalizing certain providers. That consultation runs until Nov. 9.
Meanwhile, the United States is moving through a far less unified landscape. More than 1,500 AI-related bills were introduced in U.S. state legislatures during 2026, with 280 adopted, according to figures cited by the National Conference of State Legislatures. Disclosure requirements, deepfake restrictions and rules for automated decisions are proliferating, while federal lawmakers have not passed a comprehensive AI statute this session. The resulting patchwork is already raising liability and insurance questions.
The case for caution
The strongest argument for firm rules is not abstract fear of machines. It is the practical reality that automated systems can influence employment, housing, education, finance, health care and access to public services. In those settings, a flawed output can become a lost job, a denied loan or an opaque bureaucratic decision.
People also need to know when they are dealing with synthetic media or an automated agent. Transparency cannot solve every problem, but it gives citizens a chance to challenge decisions and assess information. The EU’s transparency obligations, including marking requirements for qualifying AI-generated content, are designed around that principle, though transitional arrangements extend some compliance deadlines into December. The implementation timetable reflects both the scale and complexity of the task.
Public opinion supports guardrails. A Quinnipiac poll cited in recent coverage found that 91 percent of respondents considered U.S. AI safeguards important, while 81 percent said safety should come before innovation. Those figures should not be treated as a mandate for any particular law, but they do undermine the claim that oversight is merely an elite obstruction to progress.
The case against overreach
The counterargument deserves equal weight. AI is developing faster than legislation, and poorly drafted rules can entrench large incumbents, raise compliance costs for small firms and discourage beneficial experimentation. A requirement written for today’s generative chatbot may be ill-suited to tomorrow’s scientific tool. Businesses also face genuine uncertainty when national and state rules overlap or conflict.
That concern is not theoretical. Colorado rescinded and replaced a landmark automated-decision regulation in 2026, while U.S. jurisdictions continued to pursue different approaches to audits, disclosures and high-impact uses. The shifting landscape shows why durable principles matter more than hurried technical prescriptions.
Yet the answer to regulatory complexity cannot be regulatory surrender. Companies already operate under obligations involving privacy, product safety, discrimination and consumer protection. AI should not receive a blanket exemption simply because its outputs are probabilistic or its capabilities are novel.
What responsible policy looks like
First, lawmakers should concentrate on outcomes and uses rather than trying to classify every model forever. A system used for entertainment should not face the same duties as one used to determine eligibility for housing or medical care.
Second, rules need independent enforcement, accessible complaint channels and meaningful remedies. Disclosure without accountability is public-relations theatre. Regulators should have technical expertise, while companies should retain clear records about training, testing, deployment and significant incidents.
Third, governments should coordinate. The EU, U.S. states and other jurisdictions need interoperable definitions and reporting expectations where possible. Singapore’s new AI risk-management guidelines for financial institutions offer one example of a phased, sector-specific approach, with core requirements scheduled to take effect in 2027 and further controls in 2028. Phasing can give institutions time to build credible safeguards.
Our newsroom’s position is therefore neither “regulate everything” nor “let innovation decide.” It is simpler: regulate the harms people can reasonably foresee, make powerful deployers answerable for foreseeable failures, and revise the rules as evidence changes. Innovation may move quickly, but public trust is harder to rebuild. The race to govern AI should be judged not by the number of laws passed, but by whether ordinary people can understand, challenge and survive the decisions these systems make.