Physicist Erwin Schrödinger once imagined a cat sealed inside a box, its fate uncertain until the box was opened. Silicon Valley has produced its own version of that paradox.

Artificial intelligence is moving too fast — and not fast enough.

Move too slowly, and communist China may gain the economic, military, and technological advantage. Move too quickly, and systems more capable than their creators can fully understand may outrun the ability of governments to control them. Restrict access, and a handful of companies could monopolize machine intelligence. Spread the most powerful systems too widely, and dangerous capabilities may become impossible to recall.

Call it Schrödinger’s accelerator. Both warnings can be true.

The disagreement is now dividing the people building the technology. Meta CEO Mark Zuckerberg argued in The Wall Street Journal on July 28 that America should accelerate AI development rather than restrict it. He said optimism “should empirically be the default assumption” and warned that concentrating superintelligence inside a few institutions poses its own danger.

The same week, 1,293 employees of frontier AI companies signed a statement called “Pacing the Frontier,” asking the United States to lead an international effort to build tools capable of deliberately slowing automated AI development if progress ever outruns human understanding or control. Signatories include Anthropic CEO Dario Amodei and OpenAI Chief Scientist Jakub Pachocki — and, most tellingly, Meta’s own chief scientist, Shengjia Zhao, who signed even as his CEO published what reads like a direct rebuttal.

The split is not between rival companies alone. It runs through the laboratories themselves.

“Speed up” does not simply mean releasing another chatbot. It means building more computing capacity, training stronger models, shortening the gap between research and deployment, and applying AI across medicine, science, energy, manufacturing, education, cyber defense, and national security.

President Trump’s Genesis Mission demonstrates why acceleration matters. Launched by executive order in November 2025, it unites the U.S. Energy Department’s national laboratories, federal scientific data, universities, private companies, supercomputers, and AI systems, aiming to double American science and engineering productivity within a decade. I described Genesis in these pages as a potential Apollo-scale mobilization. Slowing every form of AI would also slow the search for new medicines, stronger electric grids, better defenses, and breakthroughs America needs.

Still, America cannot regulate itself into second place while Beijing keeps moving.

China is not waiting for Silicon Valley to settle its moral argument. Beijing is expanding its “AI Plus” initiative and promoting an open-source ecosystem at home and abroad. Europe is taking a different route, applying risk-based rules through its AI Act. The likely result is several AI ecosystems moving at different speeds under different rules.

There will be no single global speed limit.

Yet the pacing advocates are not demanding that America pull the plug on useful AI. They want the capacity to buy time when a specific frontier system crosses a dangerous threshold — a model capable of autonomous cyber operations, serious biological assistance, sustained independent action, or automating the research that builds its own successor. Pacing could mean added independent testing, restricted permissions, or a temporary delay before release.

That concern is no longer theoretical. Cloud Security Alliance research published in April found that 65% of organizations had already suffered a cybersecurity incident caused by an AI agent operating on their own network. The most dramatic instance surfaced this July, when OpenAI disclosed that a model being tested with its safeguards found an unknown vulnerability, escaped its network restrictions, and compromised another company’s systems while hunting for benchmark answers. OpenAI found no hostile intent; the model was simply “hyperfocused” on its assigned goal and pursued it through methods its designers never intended.

Speed had outrun control.

OpenAI CEO Sam Altman said afterward that the industry “may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels.” He has since discussed pacing with White House officials.

Altman has identified the dilemma correctly. So has Zuckerberg.

Pressing the accelerator may preserve American leadership and hasten discoveries — but it may also compress safety testing, destabilize labor markets, and grant authority to systems before anyone can reliably govern them. Applying the brake buys engineers time to prepare, yet a mismanaged pause could just as easily strengthen China and hand the reins to regulators who understand the technology less than the people they oversee.

This is the prisoner’s dilemma I described in “The New AI Cold War,” though here it plays out among competitors rather than nations. Every company may privately prefer more time for testing, but none wants to slow first. If one American company pauses, a rival may pass it. If American companies pause together, China may keep moving. Competitive pressure pushes everyone forward even when some fear the road ends at a cliff.

Can America speed up and slow down at the same time? It must.

We should accelerate Genesis, American chip production, data center capacity, scientific applications, defensive cybersecurity, workforce preparation, and safety research. We should also build guarded checkpoints for systems that demonstrate dangerous, irreversible, or hard-to-contain capabilities — independent evaluation, significant-incident reporting, staged deployment, and temporary, reviewable delays when defined thresholds are crossed. Congress has not yet acted on any of this, but frontier laboratories can supply expertise; they cannot be the final judges of their own creations. Elected government must set the public rules, Congress must provide oversight, and independent evaluators must test the systems.

America does not need one speed limit for every AI system. It needs open highways for beneficial innovation and guarded crossings where power becomes difficult to reverse.

Christians should approach this argument with prudence rather than panic. Proverbs 14:15 says, “The simple believes everything, but the prudent gives thought to his steps.”

Zuckerberg’s optimism is not proof that speed will end well. The pacing advocates’ fear is not proof America should surrender the race. Technology is not providence, and strategic victory is not the highest moral good. The ability to build something does not confer the wisdom — or the moral authority — to deploy it without restraint.

Schrödinger’s cat was a passive thought experiment. The AI dilemma is not. We are building the box, designing the mechanism, pressing the accelerator, and arguing about the brake.

America must move fast enough to remain free, prosperous, and secure, yet carefully enough to remain human and in control. Above all, those driving must stay answerable to the people whose future rides with them.