Artificial intelligence moves at the speed of computing — models doubling in capability, products shipping in weeks, competitors releasing features within days of each other.
Regulation moves at the speed of parliaments — hearings, drafts, amendments, consultations, years between an idea and a law. The collision between the two speeds is the defining governance story of the decade.
The speed mismatch
The gap between technological and regulatory speed is not new, but with AI it has become unusually wide.
A model can change what is possible overnight. A regulation takes years to write and longer to enforce. By the time a rule addresses a particular AI behavior, the technology may have moved somewhere else entirely. The rule chases a moving target and usually arrives late.
This mismatch has produced two responses: those who conclude that regulation is hopeless, and those who conclude that it is essential — just different. Both have a point, and the truth lies somewhere between.
The case for acting anyway
Despite the speed gap, the case for regulating AI is not as weak as it sometimes sounds.
The first argument is that the most consequential uses of AI are not the fastest-moving ones. Credit decisions, hiring, medical diagnosis, insurance — these applications change slowly, because they sit inside institutions that change slowly. A regulation that addresses these uses may arrive on time even if it misses the latest model.
The second argument is about baselines. Even imperfect regulation establishes standards — what must be disclosed, what must be tested, who is accountable — that shape the industry even when the specifics age quickly. The baseline matters more than the details.
What regulators are actually targeting
Look at the regulation being written, and a clearer picture emerges of what governments actually fear.
It is not the science-fiction risk of machines rebelling. It is the mundane, present-day risks: deepfakes that deceive voters, automated decisions that discriminate, systems that fail without anyone being accountable, concentration of power in a few companies. The regulations are aimed at these concrete harms.
This is reassuring, in a way. The governance debate is not being conducted in the abstract; it is being conducted around real, identified risks that have already started to materialize.
The enforcement problem
Writing rules is one thing; enforcing them is another, and enforcement is where the new governance faces its hardest test.
AI systems are opaque, distributed and fast. Proving that a particular outcome was caused by a particular algorithmic choice is technically difficult. Regulators are discovering that their traditional tools — inspections, fines, legal action — are ill-suited to systems they cannot fully inspect.
This is pushing regulators toward new instruments: audits of algorithms, requirements for testing and documentation, obligations to explain decisions, and a greater role for independent reviewers. The shift is from punishing after the fact to requiring process before it.
The global patchwork
The governance of AI is developing unevenly around the world, and the patchwork has consequences.
Some jurisdictions are writing comprehensive, binding rules; others are relying on voluntary codes; still others are creating space for development with few constraints. Companies operating across borders must navigate a maze of different requirements.
The patchwork has a logic, though: it is an experiment. Different approaches will produce different outcomes, and the evidence of what works and what fails will accumulate. The risk is that the patchwork also lets companies choose the weakest rules — a race that serves nobody.
What companies should do
For companies in the AI space, the honest advice is to treat governance as a design constraint, not an afterthought.
The companies that embed transparency, testing and accountability into their products from the start will be the ones that adapt fastest as rules arrive. The companies that ignore governance until forced will face expensive retrofits and reputational damage.
There is also a commercial argument: trust is becoming a competitive advantage. As AI becomes ubiquitous, the systems that can demonstrate reliability and accountability will be the ones that win the cautious customers, the public contracts and the regulated markets.
The verdict on the speed limit
The title of this piece calls governance the new speed limit on AI, and that is true in a specific sense.
Regulation is not going to stop AI development; the momentum is too strong and the incentives too large. What regulation will do is shape which directions of development are rewarded — favoring the transparent over the opaque, the tested over the untested, the accountable over the black box.
That is a form of speed limiting, but not the kind that frustrates. It is the kind that makes the remaining speed safer. A technology that is powerful enough to reshape economies deserves to be directed with care, even at the cost of some pace.
The slow institution of regulation, meeting the fast institution of AI, is not a defeat for either. It is the beginning of a working relationship — and like most relationships, it will be messy, contested and, if handled well, productive.