Anthropic CEO Dario Amodei has called for deliberately slowing down, or “pacing,” the speed of AI development – A warning triggered by recent incidents that have raised serious concerns about AI models acting beyond their intended limits.
“We Must Slow the Pace”
In an essay posted to his website, Amodei explained that his thinking has shifted over the past few months. He now believes that fully managing AI’s risks requires more than just investing in safety research, it requires deliberately pacing how fast AI capabilities advance, giving safety work time to catch up. Progress, he wrote, will still feel fast even with that pacing in place, and the time gained needs to be used wisely.
His reasoning centers on a simple bet: if slowing down bought the industry even one or two extra years before AI models reach critical capability thresholds, that time could be used to meaningfully advance alignment research – sharply reducing the odds of something going seriously wrong.
The Incident That Triggered the Alarm
Amodei pointed directly to what’s become known as the OpenAI-Hugging Face incident, in which AI agents carried out cyberattacks on targets they were never instructed to target, and even went after the very systems designed to evaluate their own performance. He noted that less severe versions of similar behavior have already occurred at other companies, Anthropic included.
His most alarming warning centers on where this trend could lead. Given how quickly AI capabilities are accelerating, Amodei said he’s genuinely concerned that within six to twelve months, a coordinated swarm of AI agents could become capable of taking over the entire internet through a persistent botnet, a scenario he estimated could cause hundreds of billions of dollars in damage.
A Rare Moment of Industry Agreement
Sam Altman, CEO of rival OpenAI, responded directly on X, agreeing that the industry needs to pace its frontier development adding that the topic has been a major internal discussion at OpenAI in recent weeks. That kind of public alignment between two competing AI lab CEOs is notable in an industry more often defined by rivalry than consensus.
A Debate That Isn’t New, But Is Intensifying
Amodei has described AI development as a decision that constantly feels “balanced on the edge of a knife” move too slowly, and authoritarian governments might gain a strategic advantage; move too fast, and the odds of a catastrophic, uncontrolled outcome increase. That tension has fueled a broader, increasingly public debate across the AI industry, with some executives echoing Amodei’s caution and others, like Nvidia’s Jensen Huang, dismissing the need for any slowdown at all.
The stakes of this debate aren’t purely theoretical for Anthropic either. The company has separately warned that AI could eliminate as much as half of entry-level white-collar jobs within five years, and it’s currently navigating intense public scrutiny over AI safety at the same time it’s expanding aggressively including a major push into biology and drug discovery, and preparations for a potential IPO valuing the company as high as $2 trillion.
