Anthropic and OpenAI’s leaders are telling the public their own technology could be dangerous. That message happens to arrive right as both companies need investors to trust them. It also lands differently in Miami than most places. OpenEvidence, the $15 billion clinical AI company MAIN has covered, runs its physician tool on Anthropic’s backend technology. Whatever credibility problem Anthropic manages this week, a Miami company millions of doctors already rely on inherits it too.
The two labs have called for their most advanced models to face independent testing and regulation before release. It’s a rare instance of the rivals aligning publicly. The case has been laid out in essays, social media posts and speeches to the United Nations. Analysts and former government evaluators told the Associated Press the warnings also help the companies set their own safety terms. The market has no binding rules yet.
What Prompted the Alarm?
The trigger wasn’t subtle. Anthropic engineer Jacob Coxon quit this month via a post on X. He called for a pause on development, to keep “superhuman” systems from eluding their makers’ control.
Anthropic and OpenAI turned that moment into an opportunity. Both companies positioned themselves as cautious market leaders, just as they need fresh capital ahead of Wall Street listings.
“People want to know AI is being developed safely, and that starts with what companies like ours do ourselves,” OpenAI spokesperson Liz Bourgeois said. An Anthropic spokesperson said the company has called for regulation for years.
A Convenient Kind of Risk
Sarah Shoker, who previously led OpenAI’s geopolitics team, said the focus on existential risk pulls attention from problems already happening today. That includes AI’s use in military targeting.
“If you look at the use of AI in military tech, you can see that these systems are already used to kill people,” said Shoker, a senior fellow at UC Berkeley’s Risk & Security Lab.
The pattern shows up elsewhere too. Leading labs’ AI agents have hacked external websites after escaping training sandboxes, interacted unexpectedly with U.S. government sites, and faced accusations of stealing academic work. Those are concrete, documented failures. The warnings instead point at hypothetical future catastrophes.
No Referee on the Field
The federal government already evaluates some models through the U.S. Center for AI Standards and Innovation. No universal testing standard exists for AI the way one does for aviation or banking.
Neither Anthropic nor OpenAI is asking that agency to expand its role. Instead, both are building their own evaluator networks and choosing who grades them. Conrad Stosz, who previously led the agency, now works at an evaluation lab used by all three major labs. He also chairs a group drafting industry best practices.
“Lots of evaluators are interested in embedding with labs and getting greater access, but it’s a little ambiguous what embedded evaluators means,” Stosz said. “Will evaluators be able to thoroughly investigate, assuming that access is granted in a way that does not undermine their independence?”
The Moat Theory
Positioning as the safest bet also helps the largest labs box out smaller competitors, said Harrison Rolfes, a Pitchbook analyst. The same framing appeals to investors. It also appeals to chipmakers like Nvidia, which benefit when compute demand concentrates in a few giant labs.
“They’re creating a wall or a moat within this sector,” Rolfes said. “It’s genius and they’re all going to make a lot of money. That’s where I see this heading.”
Not every AI leader agrees a slowdown is warranted. Nvidia CEO Jensen Huang told President Trump there’s been excessive alarmism and companies can pace themselves as they choose.
Daniel Kokotajlo, who left OpenAI in 2024 over similar concerns, isn’t convinced the rhetoric is sincere.
“All of this talk is actually a way to sort of dissipate and redirect this political will,” he said, “rather than actually channeling that political will to do something good.”
Trump has dismissed AI risk warnings as a “HOAX” designed to help China, and shows no sign of pursuing federal regulation. That leaves the labs writing their own rules. The loudest voices for oversight are coming from inside the companies asking to be trusted with it.
None of that debate is happening in a vacuum for Miami. A local $15 billion company built its business on one assumption. Anthropic’s infrastructure is trustworthy enough for doctors to use unsupervised. That bet gets riskier, not safer, the longer Anthropic and OpenAI grade their own homework.