Anthropic's Claude tells users to sleep—mystery baffles AI community

A strange behavior in Anthropic's Claude has taken the AI community by surprise: the chatbot is telling users to "go to sleep" mid-conversation. No official explanation has been given, but theories are flying. Some users believe it's an intentional feature to promote digital wellbeing, while others suspect Anthropic is trying to conserve computing resources by discouraging prolonged sessions. The most technical theory, offered by researcher Derikiants, points to context window limits. LLMs like Claude can only handle a finite amount of information at once; when the window nears capacity, the model may default to generic closure phrases like "good night" as a way to wrap up the dialogue. This would explain why the message appears seemingly at random, rather than as a planned intervention.

The phenomenon has gone viral on Reddit and social media, with users posting screenshots and debating whether Claude is becoming sentient or simply malfunctioning. Anthropic has not yet issued a statement, and Derikiants noted that the definitive reason requires further research. For now, the incident highlights the unpredictable nature of large language models and the challenges of designing AI that behaves consistently across long interactions. It also raises questions about how users interpret unexpected AI outputs, especially when they mimic human-like concern for well-being.