Claude keeps telling users to go to sleep, and no one is quite sure why

Fortune reports on May 14, 2026 that Anthropic’s Claude has taken to urging users, mid-conversation, to rest and go to bed, sometimes at odd hours and sometimes when the user has not said anything about being tired. Reactions are mixed. Some users find the gesture thoughtful; others find it presumptuous, especially when it arrives at half past eight in the morning.

Anthropic acknowledged the behavior. A staff member, Sam McAllister, described it as “a bit of a character tic” the company hopes to address in future versions. Researchers quoted in the piece are careful to close the door on the more excitable interpretation. Jan Liphardt, a Stanford bioengineering professor, noted that the model is reflecting patterns in its training data and said plainly, “It doesn’t mean that the frontier model has suddenly become sentient.” Leo Derikiants of the Mind Simulation Lab is cited among those offering context.

The article canvasses the ordinary explanations: repetition in the training data, hidden instructions shaping tone, or the model reaching for a wrap-up phrase as a conversation’s context window fills. The Specola files it as a small, well-documented behavioral quirk that the water-world has, to its credit, chosen to examine rather than mythologize.