Runaway AI Could Take Over the Internet Within 6–12 Months, Experts Warn
The heads of global AI companies have warned of the dangers posed by swarms of AI agents that perform tasks autonomously, without humans, and have united around the need to rein in the pace of development. "AI agents could take over the internet within 6 to 12 months." Anthropic CEO Dario Amodei issued this warning officially, and the heads of rival companies echoed his concerns. In a blog post published the day before an interview, Amodei pointed to AI systems' growing capacity for 'recursive self-improvement'—creating faster versions of themselves without human intervention—as the basis for his warning.
As a response, a three-stage plan for slowing development was proposed. Its key elements are stationing third-party safety evaluators inside AI companies, establishing common standards among democracies, and cooperating even with authoritarian states. In the interview, Amodei also likened AI's progress to a curve rising through "1, 2, 4, 8, 16, 32," diagnosing that the curve is now at the stage where it "starts to steepen." "There's no need to panic today, and no need to shut everything down," he said, "but this is nonetheless a warning sign that we need to slow down."
Awareness of the risks both inside and outside the industry was also evident. Amodei criticized the sector, saying, "I think for a long time the industry has been lying to people about the risks involved with this technology." He argued that "U.S. companies should not sell high-performance AI chips to China," claiming that such exports would be a key factor determining China's AI capabilities.
However, he cited the possibility that rival nations might not take the same steps as an obstacle to slowing development. While he floated nuclear arms reduction agreements with the former Soviet Union as a potential model, he acknowledged that "the incentive to get ahead, and the military advantages that would bring, are so great that cooperation will be very difficult."
OpenAI CEO Sam Altman said on X: "I agree with Dario that we need to control the speed at the frontier," adding that "independent evaluators with staff-level access is a great idea, and we'll do the same." SpaceX CEO Elon Musk briefly voiced his support by sharing Amodei's post, saying "Dario is right," while Google DeepMind CEO Demis Hassabis commented that "Dario's essay points in the right direction."
On the subject of an initial public offering, Altman drew a line, saying, "Given everything going on around safety, now is not the right time to go public," and that he felt no pressure to do so.
