Anthropic CEO calls for pacing AI frontier model development and warns of potential internet takeover

Anthropic CEO Dario Amodei has released a comprehensive proposal advocating for the strategic 'pacing' of frontier AI model development. In a 3,000-word blog post, Amodei warns that without proper oversight, advanced AI systems could pose significant risks, including cyberattacks and potential loss of control. He specifically cautioned that within 6 to 12 months, autonomous agent swarms could become capable of taking over the internet. His framework suggests implementing independent internal ombudsmen at AI companies, establishing standardized evaluation checkpoints, and fostering democratic coordination among global leaders. While Amodei emphasizes the need for caution, he argues against a total pause, noting that maintaining a competitive lead over autocracies is essential for national security. The proposal also touches on restricting AI chip exports to China and implementing penalties for model distillation to ensure that the rapid evolution of AI remains aligned with human safety and global stability.