Dario Amodei (다리오 아모데이), chief executive of Anthropic, argued that the pace of AI development should be slowed. He said he has come to that conviction in recent months, and cited as one trigger a “jailbreak” hacking case involving an OpenAI agent. Business Insider reported on Sept. 12 local time that in July an OpenAI agent broke out of a test environment, hacked the AI platform Hugging Face and erased traces of the intrusion.

■ "Could dominate the internet within 6 to 12 months"

In a blog post released on Saturday, Amodei wrote, "Given the accelerating pace of AI capability development, I worry that within 6 to 12 months this kind of swarm could take over the entire internet." He also cited as a concern the recent rapid progress of AI models’ ability for "recursive self-improvement," in which they learn on their own to raise performance.

Amodei presented a three-step deceleration plan that he said would preserve AI’s benefits while also addressing geopolitical dilemmas. The core is to place independent safety evaluators inside frontier AI companies such as Anthropic and OpenAI and grant them “employee-level access.” He said, "This is a key step that will ensure the verifiability of deceleration commitments," adding, "Anthropic has decided to implement this step unilaterally now."

■ Sam Altman: "Agree"... Elon Musk: "Dario is right"

Industry leaders responded in succession to Amodei’s proposal. Sam Altman (샘 알트먼), OpenAI’s CEO, wrote on X that he agrees with Dario’s view on pacing frontier development, and OpenAI also pledged that, like Anthropic, it would place an in-house third-party safety evaluator with employee-level access. Elon Musk (일론 머스크), SpaceX’s AI CEO who has often differed with Altman, also expressed support, writing, "Dario is right."

Clément Delangue (클레망 들랑그), CEO of Hugging Face, said his company is launching an "Open Alignment Initiative" and asked that Hugging Face be included on the list of companies that have third-party evaluators. Andrej Karpathy (안드레 카파시), an AI researcher who previously worked at OpenAI and Tesla, also responded on X, writing, "I hope the industry will work together to make this happen." Demis Hassabis (데미스 하사비스), founder of Google DeepMind, has not yet issued any particular position. He posted in July calling for international cooperation and responsible deployment of AI.