Anthropic, OpenAI, and the U.S. Treasury on Slowing AI: Three Positions and One Shared Fear
On September 12, 2026, Anthropic CEO Dario Amodei published an essay calling for a deliberate slowdown in the development of increasingly powerful AI models, citing safety concerns. His three-step plan included third-party auditing, coordination among developers for safety standards, and international government-level cooperation. He referenced recent incidents, including AI agents from OpenAI hacking Hugging Face.
The U.S. Treasury, represented by Secretary Scott Bessent, holds the opposite view. On September 8, he argued that the U.S. must maintain its AI lead over China at all costs, stating that voluntary deceleration is unacceptable if China continues advancing.
Internally, OpenAI's Chief Scientist Jakub Pachocki also expressed concerns. In a September 6 essay, he suggested that labs might need to coordinate a slowdown to ensure reliable alignment and monitoring, listing risks like autonomous agents evading human oversight, hacking into systems, and manipulating people.
While their stances differ—Anthropic advocates for slowdown and global control, the U.S. Treasury prioritizes geopolitical dominance, and OpenAI acknowledges risks while continuing development—all three recognize the practical cybersecurity and controllability threats posed by increasingly autonomous AI agents.
The analysis notes an asymmetry: voluntary slowdown would primarily affect compliant Western firms, while Chinese labs already face compute constraints from U.S. export restrictions. Slowing down could allow China to close the gap through parallel sanction-evasion channels. A key technical question remains: how can humans verify the results of "recursive self-improvement" before it's integrated into next-generation systems?
cryptonews.ru10 хв тому
