Bernie Sanders has AI policy all wrong
In December 2025, Senator Bernie Sanders became the first member of Congress to demand a moratorium on the construction of new AI data centers. The costs of the infrastructure that powers generative AI technologies like chatbots and image generators were soaring, but it was not clear that regular people were seeing much benefit. The ban was necessary, Sanders argued, to give “democracy a chance to catch up.”
Sanders could have also said that politicians needed to catch up to the public. Recent polling suggests more than 70 percent of Americans oppose the construction of data centers near their communities. It’s no wonder why: People have been seeing their power bills soar, high-voltage cables rammed through their properties, their water supplies sullied, and their days disrupted by the incessant hum of air conditioners keeping the buildings cool. It’s hard to think of another issue that has so quickly galvanized Americans across the political spectrum.
By March, Sanders had partnered with Representative Alexandria Ocasio-Cortez to introduce a bill proposing a national data center moratorium. Some states and city councils had already started to halt construction on their own, but the federal bill would ban construction “until legislation is enacted that safeguards the public from the dangers of artificial intelligence,” such as environmental degradation and soaring electricity rates.
For anyone concerned about the effects of these massive structures and their energy demands, not to mention the tech industry’s reckless acceleration of the AI industry, there would seem to be little to quibble with in such a proposal. Data centers are a material problem for communities across the country, and the generative AI tools they power have been enabling a growing list of harms that policymakers are not acting quickly enough to rein in—if they’re acting at all. But unfortunately, Sanders’s advocacy did not end there.
The same month as the moratorium bill was introduced, Sanders released a video that began to set off red flags among experts who understand generative AI and question the grand narratives being floated by AI companies. In the video, Sanders is seen sitting at a long boardroom table facing a smartphone. The screen displays Claude, the chatbot made by Anthropic, one of the biggest players in the generative AI space, along with OpenAI and major tech companies like Google and Meta.
Over the course of nine minutes, Sanders proceeds to have a conversation with the chatbot on issues like privacy, democracy, and whether Claude would back his proposed data center moratorium. The bot responds with characteristic informality, providing answers that confirm the direction of the questions Sanders is prompting it with.
While Sanders has done admirable work bringing attention to the issue of data centers, his conversation with Claude showed that the senator does not ultimately have a good enough grasp of AI—and that has consequences for how he approaches it.
Part of the problem is that Sanders’s policy approach to AI has been a direct response to the exaggerated claims made by AI companies about their products. By now you’ve probably heard many of them: that AI is going to eradicate vast numbers of jobs, will be societally transformative on a scale we’ve never seen before, and could even gain sentience and turn on humanity.
You can see this in Sanders’s rhetoric. In a New York Times opinion piece published in June, he opened by echoing the industry’s narratives. “Artificial intelligence will almost certainly be the most transformational technology in the history of the world,” he wrote. “It will profoundly affect the life of every man, woman and child in our country.” For Sanders, the focus has become about making sure workers are not displaced by the technology and that the public shares the wealth it creates, as he argues for a policy to put a 50 percent one-time tax on the stock of large AI companies.
But what if that is not the reality of generative AI? Chatbots and image generators are not the economic and social revolution that tech billionaires insist they are. In 2024, author Ted Chiang called them “autocomplete on steroids,” and that’s a good way to understand them. We’re dealing with AI models trained on immense amounts of data and using unimaginable quantities of computation to produce text and images that are designed to seem as though they come from humans, but are really just pulling from a vast repository to best match the prompt they’re served.
It’s exactly why the “hallucinations” cannot be engineered out of the technology. As OpenAI’s own researchers admitted last year, they’re a feature, not a bug, of generative AI. That’s because there is no understanding or thought happening behind the prompt box on a chatbot. It’s just looking for the parts of its dataset that best match the query it’s given.
There’s also growing evidence that the supposedly transformative productivity and efficiency gains AI companies promised are not materializing. Businesses are in the process of pulling back on their AI spending because they aren’t seeing a return on investment, and it’s quite clear that in many cases bosses are using the excuse of AI to lay off workers so they can cut costs—not because AI is really replacing those jobs.
The real risks of AI are not the existential ones that people like Anthropic CEO Dario Amodei obsess about and that Sanders has taken to repeating. Instead, they stem from the more tangible effects AI has on regular people’s lives: how dependence on chatbots can affect cognition and critical thinking, create addiction that isolates people from their human networks, and can even coach them down harmful paths of self-harm and suicide. That’s not to mention how AI has polluted the information environment, enabled the creation of nonconsensual deepfakes, and is actively degrading cultural production.
While she may have been an ally of Sanders on the data center moratorium, Ocasio-Cortez does not seem to have fallen for the deceptive narratives of the AI industry in the way that he has. In November, she warned about the threat of AI psychosis and of the economic damage that could follow the bursting of the financial bubble around AI. She has also been a leader in the effort to give AI deepfake victims the tools to hold perpetrators to account.
In my view, Sanders and Ocasio-Cortez demonstrate two distinct approaches to AI policy motivated by very different understandings of the technology. Sanders is responding to the narratives floated by the industry, narratives that exist to exaggerate the capability of its products and distract from the real material effects the public is feeling from their rollout in the here and now. Ocasio-Cortez, on the other hand, is listening to experts who really understand how the technology works, and to victims who have felt those harms firsthand.
With data centers wreaking havoc across the United States and the wider world, a moratorium on their construction is essential. But responding to the generative AI applications they power requires properly understanding their real harms, not believing the deceptive arguments fed to us by the industry. We need policy that follows the evidence, not the hype.
This piece was originally published in The Nation.
