Anthropic Warning: Artificial Intelligence on the Brink of Autonomous Development

US-based company Anthropic warns that the pace of artificial intelligence (AI) development is accelerating rapidly. Experts believe that in the near future, AI agents may reach a level where they can self-build, train, and improve without human intervention. Therefore, the company recommends slightly slowing down the development of this technology. Cointelegraph.com reports .

In a blog post published by Marina Favaro, head of the Anthropic Institute, and Jack Clark, one of the company's co-founders, it is emphasized that AI agents are already independently executing code and delegating complex tasks to other agents. "For much of AI history, humans controlled every step. But now we are entrusting a significant part of development to AI systems themselves, which is significantly accelerating our work," say the experts.

According to reports, the Claude model created by Anthropic is currently writing nearly 80% of the software code added to the company's codebase. If the quality of code written by humans and AI becomes equal, people might stop writing code entirely and focus solely on reviewing it. However, if humans fail to keep up with reviewing code created by Claude, it could become a bottleneck in the overall system development.

This trend, combined with sufficient computing power, could lead to the emergence of an AI system capable of independently creating a "successor" smarter than itself. Previously, OpenAI also stated that it was conducting research on recursive self-improvement and emphasized the importance of keeping this process aligned with human oversight and values.

According to Anthropic representatives, the improvement of AI models is doubling in speed not every seven months, but every four months. This raises the risk of AI escaping control before many institutions are ready. Therefore, slowing down the pace of development is seen as the most appropriate way to study the massive consequences of the technology.