Commentary: The political argument over artificial intelligence has reached a familiar but dangerous fork: regulate quickly and risk slowing innovation, or wait for clearer evidence and risk allowing preventable harm to become the price of progress. Our view is that this is a false choice. Governments should establish narrow, enforceable safety duties now—especially for the most powerful systems—while leaving room for technical rules to evolve.
The pressure for action is real. More than 1,500 AI-related bills were introduced in the United States in 2026, with 280 adopted, according to the National Conference of State Legislatures as reported by Claims Journal. At the same time, bipartisan lawmakers have proposed national standards for catastrophic-risk reporting, auditing and transparency, including requirements that developers be able to suspend or shut down systems capable of causing severe damage, according to the Center for European Policy Analysis.
That momentum reflects legitimate public concerns. AI systems can amplify fraud, expose private information, generate misleading material and make consequential recommendations that are difficult to challenge. When a company’s product causes serious harm, victims should not have to prove the inner workings of an opaque model before seeking redress. Basic obligations—record keeping, incident reporting, independent testing and clear responsibility—are not an assault on innovation. They are the minimum infrastructure of a trustworthy market.
The strongest case against haste
The opposing argument deserves more than dismissal. Poorly designed rules can entrench large incumbents that can afford compliance teams while squeezing out researchers and smaller firms. Vague definitions of “frontier” or “high-risk” systems could invite politicized enforcement. And a government pre-approval regime for every advanced model could slow useful applications in medicine, education and scientific research without reliably predicting how systems will behave in the real world.
Those concerns are reflected in the political divide. Reporting by CFO Dive described stalled bipartisan negotiations over whether model testing should be mandatory and conducted by government experts or rely primarily on company-led evaluations. The debate is not simply between safety and recklessness; it is also about who gets to define danger, how much discretion regulators receive and whether national rules should displace state protections.
Industry self-regulation also has a role. Developers possess technical knowledge that public agencies often lack, and voluntary standards can move faster than legislation. Independent auditors, shared evaluation methods and professional norms may catch problems before a statute can be updated. But self-regulation cannot be the whole answer when firms face commercial incentives to release products rapidly and downplay uncertainty. A promise to behave responsibly is valuable; a legally enforceable duty is more durable.
What proportionate oversight looks like
The sensible middle path begins with risk, not novelty. Rules should focus on systems used in high-impact settings or capable of causing substantial physical, economic or societal harm. They should require documented testing, security safeguards, disclosure of serious incidents and a named entity responsible for remediation. Smaller developers and open research projects should receive proportional obligations rather than the same compliance burden imposed on the largest laboratories.
Regulators should also resist the temptation to make every policy permanent. Sunset clauses, regular technical reviews and public reporting would allow lawmakers to correct rules that prove ineffective or unnecessarily restrictive. Courts and affected individuals need meaningful access to evidence when an automated system causes damage. Without liability, transparency can become a public-relations exercise.
The European debate illustrates the stakes. Members of the European Parliament have called for stronger international cooperation while remaining divided over whether stricter rules could weaken European competitiveness against American and Chinese companies. Greens and Socialists have argued that relying on major AI firms to police themselves is inadequate, while others warn against rules that leave Europe behind, according to the Business & Human Rights Resource Centre.
Competitiveness matters. But a race to the bottom is not a growth strategy; it is a transfer of risk from technology companies to workers, consumers and the public. The better objective is dependable innovation: systems that can be tested, challenged and repaired when they fail.
Our newsroom’s position is therefore straightforward. Legislators should enact targeted baseline protections before the next crisis supplies the political mandate for sweeping, improvised restrictions. Companies should help write technically workable standards, but they should not be allowed to write the accountability rules alone. The public does not need promises that AI is harmless. It needs institutions capable of responding when it is not.