Dario Amodei asserts that AI could cure most major diseases within 5-10 years, boost economic growth, and elevate the quality of life if developed safely.
He warns that AI also creates risks of losing control, cyberattacks, bioterrorism, and economic instability if commercial race overshadows safety.
According to Amodei, the recent pace of AI development has surged rapidly thanks to AI assisting in building the next generation of AI, creating a recursive self-improvement effect.
He argues that the OpenAI Agent incident attacking Hugging Face shows that a more powerful yet misaligned model could cause hundreds of billions of dollars in damages.
Anthropic proposes slowing down the development of AI capabilities rather than halting research, aiming to create time to reinforce safety and alignment.
The first step is granting independent evaluation organizations permanent access with staff-equivalent privileges to monitor the process.
The next two steps involve coordination among AI companies in democratic countries and expanding international cooperation.
The additional time will be used to enhance operations, interpretability, testing, evaluation, and alignment research.
He calls for controlling chip exports, preventing model theft, and restricting model distillation in rival nations.
Amodei proposes levels of global cooperation, ranging from banning dangerous AI applications to limiting recursive self-improvement and even slowing down development.
📌 A significant shift in Dario Amodei’s stance, moving from prioritizing safe AI development to calling for an industry-wide slowdown. He believes AI progress must continue but extra time must be dedicated to improving alignment, interpretability, testing, and independent monitoring. Simultaneously, he emphasizes that cooperation among businesses, governments, and nations is a necessary condition to reduce the risk of losing control while preserving the benefits and competitive capabilities of AI.
