Anthropic Urges Global AI Pause Over Self-Improvement Risk

AI safety lab Anthropic issued a call on June 6, 2026, for a verifiable global pause on frontier AI development ^[1]. The company reported that its Claude model now generates more than 80% of its own codebase ^[1][3]. This milestone suggests the system is approaching a threshold for autonomous self-improvement ^[1].

Anthropic researchers warn that recursive self-improvement could soon outpace human oversight ^[1][2]. The proposed pause aims to allow international regulators time to establish safety frameworks ^[2]. Experts note this move follows growing concerns about the speed of AI evolution ^[2].

While Claude continues to assist in its own development, the lab emphasizes the need for collective action ^[1]. Reports indicate that this shift marks a significant moment in the history of computer science ^[3]. Without a pause, the company suggests that the risks of unmonitored growth may become unmanageable ^[1][2].