On September 10, two days after researcher Jacob Coxon, who had recently quit his job at Anthropic, sounded the alarm about the industry potentially bringing about the end of humanity, the New York City Democratic Socialists of America posted a carousel of images on Instagram. The first slide was a drawing of an ouroboros—a dragon eating its own tail—set against a hot pink background. “AI alarmism and AI hype are the same story: both keep your attention on a science-fiction future to distract you from the real harms experienced in the present,” the image reads.

The post was authored by the NYC DSA’s Tech Action Working Group, which focuses on tech regulation issues for the chapter, the largest in the country. It garnered more than 25,000 likes on Instagram. But not everyone agreed with the post: “Horrible and dangerous take,” one comment reads. “Sometimes sounding the alarm is justified.”

As politicians scramble to understand AI safety, an ideological split is opening on the American left. Among the questions it’s centered on is whether to believe tech companies’ claims that their products are so powerful that they could end the world.

Some on the left question the value in elevating theoretical concerns about AI over actual harms. They believe repeating claims promoted by the frontier labs about the dangers of AI empowers the AI companies, turning attention away from regulating the current harms of the technology and toward an impossible future.

“AI has become the latest tool of capitalist exploitation. The damage that unregulated and misused AI is already causing is clear across the country,” Alan Deng, a member of Tech Action, tells WIRED. “As socialists, we don’t trust these corporations to craft their own regulations.”

But others are following the direction of Senator Bernie Sanders, the unofficial leader of the American left, who has enthusiastically taken up the AI safety issue, asserting that AI has the potential to eliminate humanity. They say that it’s possible to address all harms from AI—both those happening now and those that could happen in the future—at once.

“The same people concerned about frontier governance and the development of superintelligence are also addressing those current harms,” says Alexander McCoy, a Democratic strategist working on AI issues. “We should address all these other important factors. Those frontier AI abilities are true and increasing in power. They are not fake or hype.”

Over the past year, Sanders dove head-first into the world of AI safety, bringing conversations that were previously happening in Bay Area circles to Washington and his large populist base. The 85-year-old was early to both opposing data centers—a massive organizing pole for the left this year—and speaking out about AI’s potential harms, from job loss to existential dangers. His data center moratorium bill, released in March, explicitly linked pausing construction with the need to pass federal regulation to protect “the future of humanity.”

Sanders’s positioning on this issue has put him in some unusual alliances, most notably with Steve Bannon, President Donald Trump’s former adviser and a right-wing populist podcast host. It has also seemingly inspired a raft of messaging from left-wing candidates running for office across the country, many of whom have been tying broader concerns around AI to the furor over data centers. On September 9, a day before the New York DSA post, six candidates affiliated with the Working Families Party, including four backed by Sanders, released statements on the need for Washington to regulate AI to prevent existential threats to humanity.

This has come with both public and private criticism. AI researcher and tech industry critic Timnit Gebru has accused Sanders and others on the left of “taking their cues from literal doomsday eugenicist cults,” referring to the effective altruist and rationalist philosophies, which underpin AI safety conversations in Silicon Valley. And Sanders’s moves have inspired some private agita within certain left-wing circles.

“Some of us are looking at Bernie and saying, what happened?” one NYC DSA member with a tech background tells WIRED.

Rumman Chowdhury, CEO and founder of the nonprofit Humane Intelligence, understands the concerns expressed by the DSA working group. Chowdhury, who formerly led the machine learning ethics, transparency, and accountability team at Twitter, thinks that all the attention that the AI safety discourse is generating will ultimately be beneficial to frontier labs.

“[The companies] grab all the attention, time, energy, and money by spouting really extreme perspectives. But then they actually backtrack to where we sit, all while they discount all of our expertise,” she says. “I think it’s hard for normal people to think of how the mind of a venture capitalist works and these investors work. The scarier and more fantastic these CEOs make these companies seem, the more investors want in on it.”

For AI companies—particularly OpenAI and Anthropic, which aim to go public in the coming years—the attention and scrutiny on their products now could also be a boon in the long run. “Part of IPOing is sharing a lot of diligence documentation on risks and how risks have been addressed,” says Chowdhury. “This way, they can say, there is a full regulatory regime that they are part of establishing in this industry to guard against the worst risks, and that we have all established exactly what the ‘worst risks’ are.”

Pragathi Barusubramanian is a member of the NYC DSA’s Tech Action working group and a leader in the group’s anti-data center organizing, which netted a major win earlier this year by helping New York pass a one-year moratorium. Barusubramanian feels there may be a disconnect in the conversation around what exactly constitutes the most pressing AI-related harms to regulate. “When it comes down to what’s happening on the ground,” they say, the conversation about existential risk “doesn’t come up at all.” Instead, they say, “there’s an understanding that AI and data center companies don’t have our interests at heart.”

McCoy, who is advising several Congressional candidates on AI, asserts that there is no real divide between actual leaders on the left on the issue. McCoy noted that several members of the DSA, including New York mayor Zohran Mamdani, have come out with recent statements about Coxon’s warnings and the Hugging Face hack.

McCoy says Gebru and similarly-minded critics are “marginal, unrepresentative, and also, frankly, picking an unnecessary distinction. It’s yes and. We should address harms like surveillance and Flock cameras, and other important factors. These do not come in any way at the expense of governing frontier AI capabilities.”

Phil Aroneanu, executive director of Irreplaceable, a left-leaning pro-AI safety organization, says that his time working as a climate activist has informed his approach to the AI regulation question. He believes that addressing both AI’s immediate and existing harms as well as future threats is both needed and possible. (McCoy helped found Irreplaceable with an initial seed fund from the Center for AI Safety.)

“This very similar intersection between the direct impact of fossil fuel extraction, people’s homes getting flooded out because of climate-change-fueled storms, and then the broader movement focusing on the longer-term issue of risks and emissions,” he says.

Sanders and New York representative Alexandria Ocasio-Cortez don’t appear to be on the exact same page, either. While she cosponsored Sanders’ data center moratorium legislation, including the language around AI threatening humanity in her own House version of the bill, she has not participated in the high-profile AI safety discussions that he has been present for. While Sanders has called AI “the most transformational technology in the history of the world,” much of Ocasio-Cortez’s language around AI thus far has focused on data centers and the harms to jobs the technology could create, and worries that the industry is creating a bubble.

When asked by a reporter about AI safety last week, Ocasio-Cortez threaded a careful line, saying that “two things can be true at once,” noting AI “can be very dangerous when pursued without any guardrails,” and warning that “we could be seeing an economic bubble collide with a technological frontier.”

As the safety conversation quickly coalesces in Washington and in statehouses around the country, the ultimate question for the left is not whether to support AI regulation, but what regulations the movement should put its time and resources behind. Dang told WIRED that the DSA Tech Working Group supports “comprehensive AI and tech regulation at the federal, state, and city level, addressing frontier model regulation while also fighting back against the ways that corporate AI is already being used.”

A convention resolution currently before the DSA, meanwhile, proposes organizational support for Sanders and representative Greg Casar’s bill to ban superintelligence altogether—hardly something the frontier labs would agree on. And any ideological concerns within the DSA itself may not matter too much when voters get a say. A Politico poll released Wednesday shows that two in three Americans think that there is at least some risk that AI could “destroy humanity.”

Barusubramanian says that, if nothing else, they’re “happy that everyone is talking about AI now.”