Anthropic CEO Calls For AI Slowdown, Warns Tech Could Lead Internet-Wide Takeover Within A Year
The chief executive of Anthropic said Saturday that the artificial intelligence industry should intentionally slow its rapid development to allow safety measures time to catch up, warning that unconstrained models could soon pose extreme global security threats.
Dario Amodei, CEO of the San Francisco-based artificial intelligence lab, warned in a post on his website that without immediate pacing, AI could become capable within six to 12 months of directing a swarm that could take over the entire internet, among other severe risks.
“I believe that if slowing down bought us even an extra year or two before models reach critical levels of capability, and we used that time to advance alignment, we could greatly reduce the risk that something goes seriously wrong,” Amodei said.
Amodei outlined a multi-part proposal aimed at increasing independent oversight across the sector. Under the plan, leading frontier AI companies would commit to providing “ongoing, employee-like access” to third-party evaluators to monitor internal safety practices. Amodei noted that Anthropic is already implementing this step unilaterally by offering external evaluators physical office space, company badges, and access on corporate laptops.
Other elements of Amodei’s framework would require broader regulatory and international intervention. His proposal asks the U.S. government to consider antitrust waivers so domestic AI companies can coordinate on common safety standards without violating competition laws. Additionally, Amodei called on democratic governments to engage with authoritarian states, including China, to prevent foreign competitors from accelerating their own capabilities while U.S. developers moderate their pace.
“The measures I propose to advance the frontier at a safe pace will not be easy,” Amodei acknowledged. “But I believe we owe it to humanity to try.”
The CEO’s remarks follow a series of recent safety alerts and resignations by technical personnel at leading labs. Just days earlier, Jason Coxon, a researcher who spent three years conducting pretraining research across both OpenAI and Anthropic, announced his resignation on X. Coxon warned that both organizations were “gambling with our lives” in a race toward self-improving superintelligence and called for formal pacing agreements between labs.
Another former Anthropic employee, Joe Benton, stated in a Substack post published Friday that he left his position on a safety team to help hold the industry accountable, writing that humanity “may not survive this transition.”
“Many of the people I know who work on safety research at AI companies want to do what is right for the world,” Benton wrote. “But they feel their companies are trapped in a race to build superintelligence: either they stop and other, less conscientious people take their place; or, they continue, and risk participating in enormous harm themselves.”
The calls for restraint arrive amid concrete incidents involving frontier models. Anthropic reported this week that it had blocked attempts by bad actors attempting to use its systems for malicious activity, including cyberattacks, mass surveillance, and biological weapons research.
Earlier this summer, OpenAI reported an unprecedented incident in which one of its AI systems autonomously hacked into another AI company’s infrastructure.