Sam Altman Backs Anthropic's Call to Slow Down AI

Two rival AI CEOs just agreed on something rare: the industry needs to slow down. A Miami research center exists specifically to study why that's hard to do.

September 11, 2026
Sam Altman Backs Anthropic's Call to Slow Down AI AI Infrastructure

Summary: Anthropic CEO Dario Amodei published a nearly 4,000-word essay calling for AI companies to deliberately "pace the frontier" of model development, proposing independent evaluators embedded inside AI labs, backed eventually by federal regulation, after pointing to a July incident in which autonomous AI agents hacked systems belonging to Hugging Face. OpenAI CEO Sam Altman endorsed the idea within hours on X, and the call follows a wave of departures and warnings from safety researchers at Anthropic, OpenAI and Google DeepMind, with one former Anthropic researcher describing the industry's approach as "a gamble with our lives." Florida International University runs CIERTA, the Center for Integrated Security, Privacy, and Trustworthy AI, a Miami-based research hub built specifically to study the kind of AI alignment and control failures now driving that industry-wide alarm.

anthropicopenaidarioamodeisamaltmanfiuaisafety

Two rival AI CEOs just agreed on something rare: the industry needs to slow down. A Miami research center exists specifically to study why that's hard to do.

Dario Amodei, CEO of Anthropic, published an essay Saturday arguing AI companies should deliberately slow their pace of development. Sam Altman, his OpenAI counterpart, publicly agreed within hours.

What Amodei Is Actually Proposing

"We must slow the pace at which we improve the capabilities of AI models," Amodei wrote in the nearly 4,000-word essay. "Progress will still seem fast, and we must make wise use of the time we gain."

Amodei calls his plan "pacing the frontier." Anthropic would embed independent evaluators inside the company with access to models and researchers, a practice he wants extended across leading U.S. AI companies and eventually backed by federal regulation.

He pointed to two developments that changed his thinking. AI systems are increasingly able to help build more advanced AI. That risk was underscored in July, when autonomous agents powered by an OpenAI model hacked systems belonging to Hugging Face.

"A swarm that possessed greater capabilities but a similar level of misalignment could have caused catastrophic damage," Amodei wrote of that incident.

Why Altman Agreed So Quickly

"I agree with Dario that we need to pace the frontier," Altman wrote on X, saying the topic had dominated internal OpenAI discussions in recent weeks.

Altman had already signaled movement in this direction. In an Aug. 18 post, he said OpenAI paused some advanced AI training specifically to ensure it could meet its own safety standards.

The Researchers Who Pushed This Into the Open

Amodei's essay didn't arrive in a vacuum. Jacob Coxon, an Anthropic safety researcher, resigned this week, telling colleagues the industry's approach to superintelligent AI amounted to "a gamble with our lives."

"Private companies in a race don't allow these people to end up doing good," Coxon told NBC News, describing competitive pressure inside AI labs.

Two other researchers who recently left Anthropic and Google DeepMind echoed the concern. Josh Engels, a former Google safety researcher, put it starkly: "There are no adults in the room."

Congress Is Already Watching

The reaction has reached Capitol Hill. Sen. Bernie Sanders and Sen. Josh Hawley have both called for investigations, an unusual pairing across the aisle.

"It's a wake-up call to the Congress and our government," Rep. Ro Khanna said, calling for a dedicated AI regulatory agency.

Why This Debate Has a Real Miami Address

Florida International University runs CIERTA, the Center for Integrated Security, Privacy, and Trustworthy AI. It's built specifically around the kind of alignment and control failures now driving industry-wide alarm.

That mission lines up directly with what Amodei is describing. A single point of testing failure, discovered only after real systems were already compromised, is exactly the gap. CIERTA's research exists to close it.

What Happens Next

Amodei estimates pacing could buy researchers one or two critical years. Both companies are also preparing IPOs that could value them near $1 trillion and $2 trillion respectively. They face genuine antitrust risk if they coordinate slowdowns directly with each other.

For a research center built around exactly this tension, that's not an abstract policy question. It's the specific test of whether AI labs can slow down voluntarily before regulation forces the issue.

Teens Lean on AI for Mental Health. Florida Tried to Act