Commentary. The argument over artificial intelligence regulation has entered a familiar and dangerous phase: policymakers announce ambitious safeguards, industry warns that innovation is at risk, and governments respond by delaying, simplifying or fragmenting the rules. Our view is straightforward: the answer is not less regulation, but better regulation—focused on demonstrable harms, independent testing and consequences that companies cannot negotiate away.
The European Union’s latest digital simplification package illustrates the tension. The Council approved changes that postpone some obligations for high-risk AI systems while adding new prohibitions, including measures aimed at “nudifier” applications and child sexual-abuse material, and strengthening the role of the EU AI Office. The package keeps a risk-based framework, but its delays show how quickly regulatory ambition can collide with implementation concerns.[EU AI regulatory update]
There is a legitimate case for flexibility. Rules written before technologies are deployed can become obsolete, especially when smaller firms face compliance costs that large companies can absorb. Developers also argue that unclear requirements may discourage beneficial uses in medicine, education, scientific research and climate adaptation. The European approach is right to distinguish between low-risk tools and systems that influence employment, credit, public services or safety.
But flexibility becomes a problem when it means indefinite postponement or voluntary compliance. A system used to screen job applicants, determine access to housing or generate medical recommendations can harm people even if its developer acted without malicious intent. The burden should not fall on individuals to discover an opaque system’s error and fight a well-funded company for redress.
Promises are not governance
The United States offers a different warning. Federal lawmakers have continued debating frontier-AI safeguards, but the Center for Democracy and Technology reported that no federal AI bills had passed during the 2026 legislative session, even as states pursued measures on automated decisions, audits and model risks.[CDT legislative update] That patchwork may produce useful experimentation, but it can also leave basic rights dependent on geography.
Industry proposals deserve consideration, particularly where they commit companies to testing, transparency and cooperation with public authorities. Yet self-regulation cannot be the foundation of public safety. Firms have incentives to release systems quickly, minimize disclosures and define risks narrowly. Independent stress tests, incident reporting and legal liability are therefore not anti-innovation measures; they are the conditions that make innovation trustworthy.
That does not require banning advanced AI or freezing data-center construction everywhere. Such measures could slow productivity gains, raise costs and push development into jurisdictions with fewer safeguards. Local objections to data centers, however, should not be dismissed as anti-technology. Energy demand, water use, grid reliability and community consent are material issues. The public deserves to know who pays for infrastructure and who bears the environmental costs.
A practical standard
The sensible path is a compact set of enforceable obligations. High-impact systems should undergo independent testing before deployment and regular evaluation afterward. Companies should report serious incidents promptly, preserve records needed for investigation and provide meaningful explanations to affected people. Regulators need technical staff, stable funding and authority to impose penalties large enough to outweigh the commercial benefits of concealment.
International coordination matters too. The United Nations climate technology process has highlighted both the promise of AI for early-warning systems and the need to help developing countries access technology and expertise.[UN climate technology update] If governance is designed only by wealthy states and dominant firms, AI may deepen the inequalities it is supposed to address.
Our newsroom does not believe every risk can be regulated away, or that every government is competent enough to write perfect rules. But imperfection is not an argument for delay without limits. The democratic test is whether people can understand when AI affects them, challenge consequential decisions and obtain a remedy when systems fail. Until those rights are real, another promise of “responsible AI” is only a press release.
Sources: EU AI regulatory update; Center for Democracy and Technology; UNFCCC.