Anthropic warns of AI self iteration risk: Claude has automatically generated 80% of company code Odaily Planet Daily News: Artificial intelligence company Anthropic has issued a warning that Recursive Self Improvement (RSI) may pose significant risks. Last week, Anthropic announced that its AI model Claude can now generate about 80% of the company's code, which the company called a key signal for AI to enter the "self iteration" stage - that the model can design the next generation of AI systems without human intervention. The concept of recursive self iteration has recently attracted widespread attention in Silicon Valley. Last month, OpenAI held a conference in San Francisco that attracted several researchers to explore the potential and risks of AI self iteration. Anthropic's statement emphasizes that with the rapid improvement of AI capabilities, it is increasingly urgent to establish monitoring and security mechanisms for potential uncontrollable behaviors. （The Information）