Anthropic has urged leading global artificial intelligence (AI) developers to consider slowing or pausing their AI model development. The company argues that as AI becomes increasingly involved in the development of subsequent AI systems, managing risks could become more challenging, necessitating mechanisms to control the pace of development.
In a blog post published on June 4, Anthropic stated, "The world should have the option to slow down or pause frontier AI development."
The company's primary concern centers on the concept of "recursive self-improvement," where AI takes on a significant role in developing future AI systems, leading to stronger AI being utilized in the creation of the next generation of AI. Anthropic believes that while this could accelerate development, it may also reduce the time available to identify and manage risks.
Anthropic explained that AI is already increasing the speed of development in the field. According to the company, its engineers are currently incorporating eight times more code per quarter compared to the average from 2021 to 2025. As of May, over 80% of the code merged into Anthropic's codebase was generated by its AI model, Claude.
The company noted that it would be ineffective for only certain companies to halt their development. If one organization slows down, less cautious companies or countries could gain an advantage, potentially lowering overall safety.
Anthropic emphasized that for any meaningful slowdown or pause to occur, major AI research institutions and companies across multiple countries must act under the same conditions. There must also be a way to verify whether development has indeed slowed and to establish the conditions under which a pause would begin and end.
The biggest challenge lies in verification. AI training tasks are easy to conceal, and the equipment and data used for training are highly versatile. This creates a strong incentive for someone to continue development secretly to gain a technological edge.
Through its research organization, the Anthropic Institute, the company plans to conduct studies to support a slowdown and pause framework. It aims to discuss the risks of recursive self-improvement and international coordination methods with policymakers, researchers, civil society, and other AI companies.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.