Commentary. Artificial intelligence regulation is entering its most consequential phase: not because governments have finally agreed on the rules, but because they have not. New York has moved to restrict AI chatbots for minors, China is preparing broader oversight of AI agents and biometrics, and India, Japan and South Korea are developing their own policy responses. Meanwhile, U.S. states continue to legislate in the absence of a comprehensive federal framework.
Our view is straightforward: the world needs a common regulatory floor, not a maze of disconnected experiments. That floor should protect people from demonstrable harms, require meaningful accountability from companies and preserve room for useful innovation. Anything weaker invites abuse; anything more rigid risks freezing a young technology before its benefits are understood.
The case for guardrails
The argument for action is no longer theoretical. AI systems increasingly influence education, employment, health care, policing and access to information. When a chatbot misleads a child, a biometric system misidentifies a person or an automated tool assists a consequential decision, responsibility cannot disappear into the phrase “the algorithm made a mistake.”
Recent policy activity reflects that concern. A November review by Simmons & Simmons identified measures involving chatbot restrictions for minors, biometric oversight, autonomous vehicles and responsible AI use in telecommunications. The range of initiatives shows that governments are responding to real-world applications rather than debating AI only in the abstract.
We support several minimum requirements: clear disclosure when people interact with automated systems; independent testing for high-risk applications; accessible routes to challenge harmful decisions; and named human responsibility when an AI tool is deployed. Companies should also document training data, known limitations and serious incidents, subject to legitimate protections for trade secrets and security.
Why fragmentation is dangerous
The present patchwork has a predictable weakness. Large technology companies can hire lawyers and compliance teams across jurisdictions; schools, hospitals, small businesses and ordinary users cannot. If one state imposes strict duties while another offers little protection, companies may face contradictory obligations and consumers may receive radically different safeguards depending on where they live.
That concern is visible in the United States. According to a November briefing from TLT citing the National Conference of State Legislatures, states introduced 1,539 AI bills in 2026, but only 280 had been enacted by early October, and just two amounted to broad legislative frameworks. Activity is not the same as coherence.
A global treaty may be politically unrealistic, and national governments should retain authority over sensitive matters such as policing and national security. But common principles are feasible. Regulators can agree on risk categories, minimum documentation, incident reporting and rights to human review while allowing countries to set stronger standards.
The counterargument deserves respect
Critics of fast-moving regulation are not all industry apologists. They rightly warn that poorly drafted rules can burden smaller developers, push research offshore and entrench the largest incumbents, which are best equipped to absorb compliance costs. Ambiguous definitions can also produce legal uncertainty, while rigid requirements may become obsolete as models and uses evolve.
Those objections argue for better regulation, not no regulation. Rules should be proportionate to risk, regularly reviewed and designed around outcomes rather than particular products. A low-risk writing assistant should not face the same obligations as software used to determine medical treatment or criminal-justice outcomes.
Regulators should also resist the temptation to treat every AI failure as evidence that the technology itself is unacceptable. Human institutions make serious errors too. The relevant question is whether an AI system improves or worsens a decision, who can detect the difference and who bears the cost when things go wrong.
A practical standard
Archange Shadows believes the immediate priority is an enforceable baseline: transparency for users, testing proportional to risk, privacy protections, human appeal and liability that reaches the organization benefiting from deployment. Public authorities should publish their own use policies and allow independent oversight, rather than demanding obligations from private companies they will not accept themselves.
Innovation needs confidence. People are more likely to use helpful systems when they know that powerful institutions cannot deploy them without explanation or recourse. The choice is not between progress and regulation. It is between responsible progress and a race in which accountability arrives only after the damage.
Sources: Simmons & Simmons on global AI policy developments; TLT AI Brief, November 2026; EY analysis of the EU Digital Omnibus on AI.