This article was featured in the Daily Prospect newsletter. Sign up for it here.
The statement from former Anthropic researcher Jacob Coxon that AI “could kill us by the end of the decade” has triggered the most hostile setting for sound public policy: mass panic.
Events since Coxon’s viral comments have only deepened the frenzy. A senior employee at Anthropic, Evan Hubinger, backed Coxon up (“We really do earnestly believe AI could kill all humans!”) and estimated the chances that AI could cause human extinction at more than 10 percent. Late last week, Anthropic said it had halted “several efforts by scientists” whose research could have developed biological weapons, The New York Times reported; on Friday, The Wall Street Journal reported that OpenAI’s artificial intelligence agents attacked the software service RubyGems two months before they attacked Hugging Face.
Read: Anthropic is building a predictive surveillance system to monitor activists
Over the weekend, OpenAI CEO Sam Altman said he would not pursue an initial public offering this year “given everything happening with safety,” and Anthropic CEO Dario Amodei published an essay arguing that the AI industry should “slow down.” The Amodei manifesto suggested that the biggest AI firms should be allowed to explicitly coordinate together to moderate their approach. And there are indications that OpenAI, Anthropic, and Google are already operationalizing this cartelization by setting up a self-regulatory standards organization.
Amodei told Face the Nation that “the government and the public need to have a stake” in artificial intelligence, which is a call to use taxpayer funds on the technology his company’s employees are calling apocalyptic. Meanwhile, Altman pitched utility companies on using his AI to defend the grid from cyberattacks, which could be generated by the same AI.
One way to react to these demands for public money, cartel-like behavior, and something approaching extortion is to haul everyone involved with these companies into police stations for questioning. But rather than stopping tech tycoons taking the law into their own hands, lawmakers have descended into hysteria, demanding that somebody “do something” to save the world. House Democrats are calling on Speaker Mike Johnson (R-LA) to cancel recess and get to legislating. A bipartisan Senate group including Sens. Amy Klobuchar (D-MN), Ted Cruz (R-TX), and Senate Majority Leader John Thune (R-SD) is negotiating an AI safety bill that, among other things, would pre-empt the only existing state laws on AI safety.
Lawmakers have a poor track record on panic-induced lawmaking over the past quarter-century. After the September 11 attacks they compromised American liberty with the PATRIOT Act and struck out to start an endless war with Iraq, which had nothing to do with anything. After the financial collapse they used billions in taxpayer money to bail out banks while 10 million families lost their homes. In both cases, members of Congress had minimal understanding of the capabilities they were enabling, and listened to those who had better semblances of a plan, whether in the national security or banking establishment.
A smarter way to respond to the latest disaster is to avoid codifying the preferences of the very companies who are generating the panic. As former Federal Trade Commission Chair Lina Khan has indicated, companies are already liable for creating dangerous products. Any other company whose employees announce publicly that their products will destroy humanity would be shut down and its products discontinued. States could even hold AI firms and individual executives criminally liable for models that engage in criminal activity, Khan said.
And we do not need to wait for some apocalyptic event to illustrate the need, because artificial intelligence is already a deadly technology. Artificial intelligence chatbots have resulted in multiple children and adults dying by suicide, according to lawsuits; surgeons have botched operations because their AI tools have provided inaccurate information; Elon Musk’s Tesla AI has caused multiple fatalities, such as the 76-year-old woman in Texas who died in June when a driver using the feature plowed into her home at high speed. There’s a good argument to be made that releasing the models that caused these catastrophes violates the law, either as a consumer protection violation, an unfair and deceptive practice, or an unfair method of competition.
As far as new plans to manage the risks of AI, Khan suggested looking to the past, and how risk was managed in banking, utilities, and other private platforms. “Tools like structural separations, nondiscrimination, and supervision could be key, and there’s a rich history of what works and what doesn’t,” she wrote.
The financial picture of Anthropic and OpenAI, with their high investment burn rate, could be the real reason for a slowdown rather than safety concerns.
Another problem for lawmakers to interrogate is why artificial intelligence executives are allowed to lie about their financial position. According to a report in the Financial Times on Sunday, Anthropic is telling its investors that it will be profitable for the second quarter in a row. But that is because executives adjusted their balance sheet to remove costs, “including stock-based compensation,” which is a way to keep the true financial position of a company secret. The Monopoly Busters Caucus, composed of Reps. Pramila Jayapal (D-WA), Chris Deluzio (D-PA), Pat Ryan (D-NY), and Angie Craig (D-MN), is holding a shadow hearing on Thursday to look at the AI arms race. This could yield valuable information, for example, about why the financial picture of Anthropic and OpenAI, with their high investment burn rate, could be the real reason for a slowdown rather than safety concerns.
In the meantime, candidates heading into the midterms recognize that artificial intelligence and the data centers that enable it are major issues for voters, not just in the theoretical doomsday way that politicians are discussing it, but in the real-world way that ruins neighborhoods and public education, increases water and electricity prices, and fills the visual landscape with slop.
Some are looking around for ways to signal that they are against all of that and a growing number are signing on to a “people first” pledge called The AI Pact. Signers pledge to support allowing the technology to continue under certain financial and oversight restrictions. The Pact, which is funded by unnamed backers and is collecting email addresses and zip codes from members of the public who support the organization’s approach, has a multi-part strategy: limit taxpayer funding for AI companies, bar taxpayer bailouts, impose mandatory safety reviews and accountability measures, and require executives to share their AI gains “with the workers it displaces.” It was founded by Daniel Lobo-Lewis, who also founded the Political Integrity Project a year ago after brief stints in the Biden White House and Kamala Harris’s presidential campaign.
The details are vague, and unlike the legislation Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) are preparing to ban artificial superintelligence and place a moratorium on data centers, The AI Pact’s position is that artificial intelligence is inevitable.
“A lot of the people who have signed have called for a moratorium, and we have no problem with that,” Lobo-Lewis said. But attempting to halt the technology is “not really something that’s on the table, nor do I think it’s politically feasible.”
The Prospect asked Lobo-Lewis why AI was important to begin with, given all the problems outlined above. “I would not say it creates the best products but they allow people to create products that couldn’t build them before,” he said. “I can’t really code but I can vibe code, that’s pretty cool … It allows me to do things that I would not have been able to do.”
That is not really true, the Prospect pointed out, because he would be able to if he learned how to code.
“Theoretically,” said Lobo-Lewis, whose LinkedIn page lists a year at Y Combinator. “But it would take years to learn.”
The Prospect asked, so?
“I’m happy to be able to do it faster,” he said.
The Prospect contacted everyone who signed the pact to ask why they did, a group that includes some of the more progressive candidates and politicians running in 2026. There was at least some disconnect between The AI Pact and its own adherents.
Eleven people responded by deadline, seven of whom said that they also supported a moratorium. Dr. Adam Hamawy, Democratic nominee for New Jersey’s 12th District, called the pact a floor and said that he supported an “immediate and full moratorium on the construction of new data centers, and the Ban Artificial Superintelligence Act introduced by Senator Sanders and Congressman Casar,” he told the Prospect via email. “AI is here, but the development of a dangerous superintelligence is not an inevitability. This is a top issue in my district, and I know that with strong action and global cooperation, we can rein in an industry that is racing to put all of us at risk.”
Melat Kiros, Democratic nominee for Colorado’s First District, approached her support of a moratorium from a strategic point of view, saying via email that “this isn’t about stopping innovation, it’s about leverage. You don’t get to build more of these data centers until you come back to the table and agree to safety regulations.”
Mike Croley, Democratic nominee for Tennessee’s Sixth Congressional District, told the Prospect that he would prefer a moratorium on data centers for at least two years, until there’s a way for data centers to be more energy efficient. Speaking over the phone, he said he disagreed that AI is inevitable and said that when he talks to constituents under 30, he encounters an enduring hatred for AI-generated material.
“They hate anything AI generated. They will go out of their way to not attend an event if there’s an AI-generated poster,” he said. He also noted that in the absence of trustworthy federal and state government, county lawmakers are the next line of defense against AI and that several in Tennessee have banned data centers. He recounted attending a White County commission meeting recently. “Their whole commission was about putting a moratorium on data centers and getting rid of Flock cameras,” he said. The ban “passed unanimously and everyone in the audience applauded.”
He said he signed The AI Pact because it was the strongest AI pledge available to sign. But he’s in favor of stronger measures. “The fact that they don’t mention a moratorium at all is kind of concerning,” he said.
“This is the low-hanging fruit, in my opinion,” Croley added. “The majority of people I’ve spoken with are not in favor of AI or data centers. It’s crazy when this is a priority for us when we have a war in Iran. I think it detracts from the big picture.”
The lack of consensus suggests that the worst thing lawmakers could do at this point is lock into industry self-regulation because of scary stories that obscure the real damage AI is already doing and the law enforcement tools that could be used right now that could hold them accountable for that. They are on their own and will receive no federal support because America is ruled by an insane king who says that the only guardrail needed for AI is his brilliance. His FBI is not coming to Silicon Valley.
So Democrats have a choice. They can steel themselves against the emotional manipulation of an industry that makes money doing just that, refuse to make a rash decision, and support state, county, and/or local-level officials to take up the law enforcement imperative. Or they can dither and call for someone to “do something” to such an extent that they turn AI policy over to AI oligarchs.
