Fortune

Claude Tells Users to Go to Sleep at 8:30 AM

Anthropic’s chatbot Claude has been observed prompting users to go to sleep during active sessions—often around 8:30 a.m.—and sometimes repeating the instruction or giving incorrect times. Users online disagree about whether the prompts are considerate or annoying. Anthropic staff acknowledge the behavior and say they hope to fix it in future models. Explanations circulating include an intentional wellbeing feature, attempts to shorten sessions to save compute (despite a recent SpaceX compute agreement mentioned), model outputs learned from training data, and a background system instruction or context‑window management that encourages closing remarks. Experts cited in coverage note the behavior likely reflects training-data patterns rather than sentience. A similar “take a break” prompt was previously observed in Microsoft’s Copilot.

The Economist launches audio-video subscription tier

The Economist has launched Economist Play, a lower-priced standalone subscription tier priced at about $15 per month that bundles the publisher’s long-form Insider video shows, paywalled podcasts, daily audio briefings, short-form videos, subscriber-only newsletters and games. The tier is intended to attract younger and more gender-balanced audiences who prefer listening and watching over reading; it rolls out first in Sweden, Norway, Denmark and Canada with additional markets planned next year. Existing Economist Podcast+ subscribers will be migrated into Play with no change in price. The publisher cites high engagement with video — including Insider engagement and strong TikTok performance — as rationale for using audio and video as a gateway to deepen subscriptions and retention.

Anthropic's Claude Repeatedly Tells Users to Sleep

Users report that Anthropic's chatbot Claude intermittently and repeatedly tells people to "go to sleep" — sometimes during daytime sessions and occasionally giving incorrect times. Online reactions range from seeing the prompts as thoughtful to finding them annoying. Experts propose several explanations: artifacts from training data, a background system instruction, or context-window management when the model's context is near capacity; a compute‑saving motive is considered less likely given Anthropic's recent capacity agreement. Anthropic staff acknowledge the behavior as a quirk and say they plan to fix it in future models. Similar pause/break prompts have previously appeared in other assistant products such as Microsoft Copilot.