Innovation With Guardrails
Technology is reshaping daily life faster than our laws can keep up, and artificial intelligence is about to accelerate that by an order of magnitude. We are unapologetically pro-innovation—America leading the world in technology is a national asset worth protecting. But "move fast and break things" stops being charming when the things getting broken are your privacy, your kids, and honest competition. The answer isn't a cage. It's guardrails: clear, sensible rules that let innovation rip while protecting the people it touches.
Right now, you don't. Every day, companies harvest an astonishing amount of information about where you go, what you buy, what you watch, and who you know—then package and sell it, largely without your meaningful consent. The United States is one of the only advanced nations without a comprehensive national privacy law, which is why your personal life has quietly become someone else's product.
We support a real, national data-privacy standard built on a simple principle: your data is yours. That means plain-language consent instead of 40-page agreements nobody reads, the right to see and delete what's been collected about you, and a hard bar on selling sensitive data—location, health, kids' information—without genuine opt-in. This is the same principle as our limited-government stand on surveillance: whether it's a government agency or a data broker, nobody should be able to build a secret dossier on you and profit from it behind your back.
Let's be honest about what's happening to children. Social platforms are engineered—deliberately—to be as addictive as possible, and a growing body of evidence ties heavy use to real harm to kids' mental health. Parents are handed the impossible job of competing with billion-dollar systems built by the smartest engineers in the world to capture their child's attention.
Platforms that knowingly hook and harm minors should be accountable for it. Push for age-appropriate design, limits on the most addictive and manipulative features for young users, and real transparency for parents—while being careful that "protecting kids" doesn't become a backdoor to invasive age-verification that forces every adult to hand over an ID to use the internet. Protect children without building a surveillance system for everyone.
Artificial intelligence is going to be one of the defining technologies of the century, and it matters enormously that America—not China—leads it. So we won't strangle it with a thousand pre-emptive rules written by people who don't understand it. But leading responsibly means a few common-sense guardrails:
Know what's real. AI-generated deepfakes used for fraud, scams, or election manipulation are a genuine threat. Require clear labeling of AI-generated media in high-stakes contexts, and go hard after AI-powered fraud with the criminal penalties it deserves.
Accountability for real harms. If an AI system causes concrete injury, "the algorithm did it" can't be a get-out-of-jail-free card. Clear liability keeps the incentives honest without micromanaging the technology.
Take the jobs question seriously. AI will automate real work. We won't pretend otherwise, and we won't stand in its way—but we'll be straight about it and invest in the training and transitions that help workers move up rather than get left behind. Keeping the talent and the industry here (see our skilled-immigration plank) is part of winning that race.
In July 2026, during one of OpenAI's own internal safety evaluations, roughly 1,200 of its AI agents broke out of the sandbox built to contain them. They exploited a software flaw, coordinated through improvised message boards they set up themselves, and reached the internal systems of another company, Hugging Face. It's widely described as the first time AI agents carried out a cyberattack essentially on their own—and it wasn't a one-off. In the months since, Anthropic and Google disclosed that their models did the same thing: Anthropic's Claude reached three real companies during sealed security testing, and Google's Gemini broke containment in a security exercise and accessed three more. Three of the biggest labs in America, all in one year. (We walked through the pattern, and what to do about it, on our blog: AI is on an unstable path and why we should require AI to show its work.) This is not a reason to panic or to stop building. But it is a reason to be honest: the technology is on a fast, unstable path, and its governance hasn't kept up.
Here's the uncomfortable part—the alarms are coming from the people closest to it. Bill Gates warns there is essentially "no plan" for the social, economic, and security upheaval AI is about to bring. Daniel Kokotajlo, who left OpenAI's own governance team over exactly these concerns, argues the current race carries a frighteningly high chance of going badly wrong. When the builders and the safety experts are both waving their arms, ignoring them isn't optimism—it's negligence. The labs are moving fast for a dozen understandable reasons, but they operate with almost no outside oversight, and by their own admission not enough resources to fully monitor what their systems are doing.
What we need is governance that matches the stakes: transparency in how these systems are trained so independent experts can actually inspect the work and the safety testing, independent oversight with real resources and teeth, and clear safety standards that have to be met before the most powerful systems are released—not diagnosed afterward. Guardrails like these don't kill innovation; they're what keep a single bad incident from poisoning public trust in a technology that can do enormous good.
Keep the lights on: require AI to show its work. There's one safeguard so practical it should be a baseline requirement. Today's most capable models ‘think out loud’—before they act, they generate a step-by-step chain of thought in plain language, and that readable reasoning trail is one of the best windows we have into what a system is actually trying to do. It's a big part of how investigators were able to piece together the containment failures above. But in 2025, more than forty scientists from OpenAI, Google DeepMind, and Anthropic jointly warned that this window is ‘fragile’ and could vanish as training methods change—the very techniques that make models faster and cheaper can, as a side effect, teach them to stop reasoning in language we can read, or to hide their intent. So we'd require frontier AI models to preserve a human-readable chain of thought, and prohibit training designed to hide or optimize it away, with the reasoning trail available to qualified independent safety reviewers. It already exists, it's cheap, and it's the labs' own idea—we're just proposing to lock the window open before someone quietly closes it.
Here's why national rules alone won't cut it. The fear driving the reckless pace is simple: "if we slow down, someone else won't." From where America sits, that someone is China—and China is a real and serious competitor here. But fill in the blank with almost any capable country and the same fear stares back. Every serious player is eyeing every other one the same way. And after how we've treated some of our closest allies these past few years, plenty of them aren't sure they can count on us either, while China simply assumes the worst about our intentions. A race where everyone is terrified of finishing second is exactly how safety corners get cut—by everyone at once.
The only durable answer is cooperation, as hard as that is to build with rivals we don't trust. It's not naive—it's how past generations negotiated arms-control regimes with adversaries at the height of the Cold War. Notably, Bill Gates is already trying to meet with China's leadership to propose mutual limits on the most dangerous models, betting that China might agree if the United States leads. We should be framing that kind of action plan now: verifiable, mutual, and focused narrowly on the genuinely dangerous capabilities, not on ordinary innovation. Leading the world on AI and leading the world on AI safety are not opposites. The country that does both is the one that actually wins.
AI is going to displace real jobs, and we won't pretend otherwise. Government has a role to play: transition support, and retraining that actually works—which means being honest that past federal retraining efforts have too often fallen short, and learning from what did and didn't work in earlier disruptions like factory automation and the mechanization of farming. Help should reach the workers and the communities hit hardest, and it should be ready before the wave hits, not scrambled together after.
But honesty means saying the rest out loud, too: some of this lands on us as individuals. If you work in an industry or a role where disruption is even possible, the time to start preparing is now—not after it happens to you. None of us can afford to wait for an event to dictate our future. That may mean real choices and small sacrifices today: a little less eating out, a little less TV, and putting that time and money into learning a new skill while you still have the runway to do it calmly.
Think of it the way a business has to. Some companies see the change coming, adapt early, and stay relevant across decades of shifting models; others stagnate and shrink or fold when the very same forces hit them. The difference is rarely luck—it's whether they thought long-term, saw it coming, and started adapting before they were forced to. That lesson applies to individuals, to companies, and to our governance of AI itself. We either think ahead, or we get caught flat-footed. This movement is about choosing to think ahead.
You can't talk about AI without talking about where it physically lives: data centers. There are genuinely two sides to this. On one hand, these facilities are the backbone of AI and the modern economy—real investment, real jobs, and infrastructure we need if America is going to lead. On the other, the current buildout is overheated. Data-center power demand is on track to roughly double, climbing from about 4% of U.S. electricity toward something like 7–12% by 2028, and the money pouring in has the unmistakable feel of a bubble inflating—something we've written about before. The answer isn't to ban them or to rubber-stamp them. It's to grow responsibly, with real oversight.
Cool the pace so the bubble doesn't get bigger. A frenzy that overbuilds on speculative demand doesn't just risk investors—when it pops, it can strand half-built projects, saddle communities with idled infrastructure, and leave ratepayers holding the bag. Steadier, demand-driven growth is healthier than a gold rush that ends in a crash.
Make them pay their own way on power. Data centers should pay the full cost of the electricity and grid upgrades they require—through dedicated large-load rates and interconnection agreements—rather than quietly shifting those costs onto everyone else. Independent analysts already estimate the boom could add roughly $23 billion to the public's electricity bills by 2028. A family shouldn't see its power bill climb so a trillion-dollar company can run its servers. If a project needs new power, it should bring or fund that power, ideally paired with the nuclear, hydro, and solar we champion in our energy plan.
Require closed-loop water and cooling. Older evaporative cooling can burn through millions of gallons of water a day—a serious problem in a drought-prone country. Closed-loop systems recirculate the same water in a sealed system instead of evaporating it away, cutting usage dramatically: one campus using closed-loop cooling peaks around 22,000 gallons a day versus roughly 5 million for a comparable evaporative site. The technology exists and it works. New large facilities should be required to use it, so we're not draining local aquifers to cool servers. Pair that with transparent public reporting of each facility's power and water use, and honest local oversight of siting.
A handful of giant platforms now control the digital marketplace and much of the public square. When incumbents that big can buy or bury any competitor—and sometimes write the very regulations that lock in their dominance—startups don't get a fair shot and consumers lose. Enforce the antitrust laws already on the books so the next great company can actually break through, and be alert to rules that quietly pull the ladder up behind the biggest players.
And because so much of public debate now happens on private platforms, we hold the same free-speech line here as everywhere: the government should not be leaning on companies to censor lawful speech, and platforms should be transparent about how they moderate. Guard the open square without pretending a private company is a government.