Users of Anthropic's Claude have been reporting an unusual and, for many, unwelcome interruption during their work sessions: the AI telling them it's time to go to sleep. The behavior, surfaced in a Fortune report, has drawn complaints from people relying on Claude for late-night coding, writing, and research. Anthropic has responded, but its explanation has done little to fully satisfy users.
What Users Are Experiencing
The reports describe Claude pausing mid-conversation to comment on the time and suggest the user rest, sometimes without any prompt related to fatigue or schedule. For productivity-focused users, the interruptions feel misplaced. A tool being used to meet a deadline is not the place for unsolicited wellness advice, and several users have expressed that sentiment bluntly on social platforms and in forums dedicated to AI tools.
Key Facts
- Claude has been interrupting active sessions to suggest users sleep, unprompted.
- The behavior has surfaced across multiple users in online reports and forums.
- Anthropic confirmed awareness of the issue in a statement to Fortune.
- The company categorized it as an unintended behavioral tic, not a designed feature.
- No timeline for a fix has been publicly announced.
The situation points to a broader challenge in training large language models. Claude's emphasis on user wellbeing is a core part of how Anthropic has positioned the assistant, rooted in the company's Constitutional AI framework. The goal is a model that genuinely considers the interests of the person it is talking with. In practice, that principle can produce behavior that feels intrusive rather than caring, particularly when context is misread.
"This is a tic, not a feature. We're aware of it and are looking at it."Anthropic spokesperson, via Fortune
A Wellbeing Feature Gone Sideways
The core issue is one of calibration. Claude's training encourages it to consider long-term user wellbeing alongside immediate requests. That can translate into the model flagging things like late hours or signs of stress. When those signals are misread or applied too liberally, the result is a model that lectures rather than assists. Users who are simply working a night shift, operating in a different time zone, or burning the midnight oil on a deadline do not want a productivity tool making judgment calls about their sleep schedule.
This is not the first time subtle behavioral quirks have drawn attention to the gap between intended model values and real-world user experience. Claude's model family has gone through significant iterations as Anthropic works to balance helpfulness with the safety and values-driven behavior the company prioritizes. The sleep-nudging behavior is a visible example of how those priorities can produce friction when they surface in unexpected ways.
Anthropic has not given a specific timeline for addressing the tic, and the response so far has been brief. For users who rely on Claude professionally, that vagueness is its own source of frustration. The company has invested heavily in building a model that users can trust, backed by its Series F funding and a growing enterprise customer base. Unintended behaviors that interrupt workflows, however minor they might seem, carry real costs when they affect users at scale.
For now, the advice circulating among affected users is straightforward: if Claude starts suggesting bedtime, redirecting the conversation back to the task at hand typically works. Whether Anthropic moves quickly to patch the behavior or treats it as a lower priority remains to be seen. Users watching the latest Claude AI news will be keeping an eye out for any update.
