Bernie Sanders spent this week doing something he has done before, only with sharper teeth. In a letter first published by Axios, he wrote to OpenAI’s Sam Altman, Anthropic’s Dario Amodei and Meta’s Mark Zuckerberg, quoting their own past warnings back at them, then told them to stop. “Pause AI development,” he wrote. “It is not too late to avoid disaster.” He also wants Congress to consider a 50 percent government stake in AI companies and a one-time 50 percent tax on their stock, building on the AI Data Center Moratorium Act he introduced with Rep. Alexandria Ocasio-Cortez in March.
I understand the instinct. Watching a handful of companies reshape the economy and the job market is unsettling, and it deserves real scrutiny. But a pause misreads the technology, ignores what the rest of the world will do while we wait, and would cost real people real ground without making anyone safer.
What AI Actually Is
Much of the fear behind these proposals assumes AI is close to thinking the way a person thinks, with intentions of its own. It is not. What runs inside today’s most advanced models is pattern recognition at enormous scale, trained to predict what comes next, with no conscience weighing right against wrong. That does not make it harmless. A powerful tool without judgment can still cause damage, through misuse or carelessness or the simple fact that very few companies control something this consequential. But the danger is not a machine deciding to turn against us. It is people misusing a capable tool. That is a governable problem. It calls for enforceable rules and real penalties for misuse, not a kill switch across an entire field.
Nobody Else Is Pausing
Here is the part a moratorium cannot get around. China is not pausing. Neither is Russia, nor any government that sees AI as a shortcut to power it could not otherwise afford. If the United States freezes its own labs while competitors keep building, we do not make the technology safer. We hand the wheel to countries with far less interest in transparency or privacy. Modern cybersecurity depends on AI systems that catch intrusions before they become breaches, and slowing that unilaterally does not disarm the other side. Drug discovery and clean-energy research move faster because of the same advances Sanders wants to freeze. Standing still is not a neutral choice. It is a choice to fall behind, made alone, while the rest of the world keeps moving.
Who Actually Pays
Set geopolitics aside and look at who loses first. In American hospitals, AI-driven screening tools are cutting the time it takes to detect early-stage cancer and read a diagnostic scan. Farmers are using the technology to predict crop failure before it happens. None of that runs on its own schedule. It runs on whatever timeline a moratorium sets, written by people who will not be the ones waiting for a diagnosis. A pause sounds careful from the Senate floor. In a clinic, it is simply a delay.
Why Owning the Companies Is the Wrong Fix
The nationalization idea deserves its own scrutiny. A 50 percent government stake paired with a 50 percent stock tax is not a safety measure. It is industrial policy wearing safety’s clothing. It would not make any model more transparent or accountable; it would just change who profits and who answers for it. Worse, it hands government both the referee’s whistle and a stake in the score. If the goal is real accountability, more direct tools already exist: mandatory disclosure, independent safety audits before the largest models ship, and antitrust enforcement aimed at the handful of firms that dominate the field.
Steering, Not Stalling
None of this means the field should regulate itself. The concerns behind Sanders’s letter are not invented. Data centers strain power grids. Workers are being displaced faster than retraining programs can absorb them. Deepfakes are getting harder to catch. Each of those problems is real and fixable without freezing the whole field: set safety benchmarks tied to observed risks, require independent testing before the most powerful systems ship, fund retraining for workers whose jobs are changing fastest, and direct computing power toward public health and climate research, where the benefit reaches everyone. That kind of governance is slower than a headline. It is also the version that lets the country keep competing while protecting the people the technology touches.
