Pausing a Dots agent doesn't stop its delegates — scheduled runs keep going

#What happened

• OpenAI launched dots on September 29, 2026 — persistent AI assistants with a cloud computer, browser, memory, and access to user-connected apps, powered by GPT-6 Astra. The first dot is included with eligible Pro or Business Premium subscriptions.
• A review of OpenAI's published Dots controls (the getting-started guide, safety explanation, and workspace management guide) surfaced an important operational limit: pausing a dot stops its main task, but work it has delegated and its recurring schedules must be stopped separately, in the Activity and Scheduled views.
• Disconnecting an app cuts off future reads but does not erase information the dot already learned. Resetting or deleting the dot removes its conversations, memories, and schedules, but cannot undo an external edit or recall a message that was already sent.
• Some actions keep mandatory gates: purchases require approval; permanently deleting data, running software from an unrecognized source, and granting new security-sensitive access require confirmation each time; password changes and money transfers are handed back to the user.
• In its Astra system card, OpenAI reports that permission-scope flags rose from 8.6% to 19.7% when five to ten tasks intervened in a chain. The company reports no severe breach or exfiltration in that test — but the chain result shows remembering that an account is connected is easier than preserving which task authorized which use of it.

#Why it matters

• Pausing ≠ stopping: managing Dots is an ongoing governance responsibility, and stopping behavior is as consequential as model capability. A "paused" dot can still have delegated tasks and scheduled runs alive, so owners must inspect and stop every part of that work.
• Disconnect ≠ memory wipe: if you disconnect a connected app, the dot stops reading new data but keeps what it already learned — privacy hygiene requires knowing that distinction.
• This is the control-surface reality of always-on agents: the launch turns agent management into an ongoing responsibility rather than a one-time setup.