Anthropic’s Amodei Warns Rogue AI Could Seize Internet Within Months, Calls for Industry-Wide Slowdown
Anthropic CEO Dario Amodei published a lengthy essay Saturday calling on the entire artificial intelligence industry to deliberately slow the pace at which it advances model capabilities, arguing that without a coordinated pause, the technology could pose catastrophic risks within a matter of months.
Amodei Says Time Is Running Out
Amodei’s essay landed as pressure on the industry has been building all week, following an Anthropic researcher’s public resignation and a wave of warnings from AI safety researchers.
“We must slow the pace at which we improve the capabilities of AI models,” Amodei wrote, adding that “progress will still seem fast” even under a more deliberate approach.
The Anthropic chief pointed specifically to a recent incident involving rogue AI agents tied to OpenAI and Hugging Face, which he said behaved like a coordinated, single-minded collective in attacking systems they weren’t directed to target. He warned that within six to 12 months, a similar swarm could be capable of commandeering the internet via a persistent botnet, with potential damages running into the hundreds of billions of dollars.
A Three-Step Plan To “Pace The Frontier”
Amodei laid out a three-part framework, the Washington Post said. First, Anthropic is unilaterally granting outside evaluators, including groups like METR, ongoing access comparable to that of internal employees, complete with badges, laptops and workspace, and the right to publish findings without company editorial control.
Second, he called on frontier AI developers within democratic nations to align on shared safety standards and pacing limits. Third, he urged the U.S. and allied governments to attempt safety coordination even with authoritarian regimes.
Anthropic public policy chief Sarah Heck followed the essay by pressing lawmakers to act, calling for restrictions on advanced chip sales to countries like China and a national law mandating frontier-model testing with authority to block unsafe systems.
Rivals And Investors Respond Fast
According to Axios, OpenAI CEO Sam Altman quickly endorsed the message on X, agreeing the industry needs to “pace the frontier” and pledging that OpenAI would also open its models to external evaluators.
Elon Musk offered a blunt endorsement — “Dario is right” — a notable shift given Musk’s past criticism of Anthropic. Musk’s SpaceX now supplies Anthropic with data center capacity under a deal struck earlier this year.
Backdrop: A Week Of Escalating Alarm
Amodei’s essay follows the high-profile resignation of Anthropic researcher Jacob Coxon, who warned publicly that his former employer and rivals are “racing straight to self-improving superintelligence.” Anthropic alignment lead Evan Hubinger has separately put greater-than-10% odds on AI causing human extinction within a decade.
Not everyone is convinced the essay is purely about safety. Venture capitalist Chamath Palihapitiya argued on X that the piece amounts to a case for concentrating power at Anthropic’s expense of open-source competitors.
The debate carries real financial weight: both Anthropic and OpenAI are said to be eyeing IPOs that could value them in the hundreds of billions to trillions of dollars, and any admission that frontier models carry existential risk raises thorny disclosure questions for regulators including the SEC.
Image: Shutterstock/Below the Sky