OpenAI has launched dots, persistent GPT-6 Astra-powered agents designed to take responsibility for ongoing work rather than wait for one prompt at a time.
Each dot receives its own cloud computer and browser, can work with approved connected apps, remembers project context, and continues making progress between conversations. It can monitor recurring responsibilities, react to supported events, and delegate background tasks to ChatGPT Work or Codex. Users can inspect its activity, redirect work, pause it, or require approval for sensitive actions.
Dots are rolling out gradually to eligible Pro, Business Premium, and Enterprise users, subject to regional and workspace restrictions. Enterprise access is disabled by default and must be enabled by an administrator.
Why it matters
Dots mark a shift from task-based assistants toward AI systems entrusted with persistent responsibilities. For builders and organizations, the important product layer is no longer only model intelligence: permissions, memory, scheduling, delegation, activity visibility, and human checkpoints now define whether an autonomous agent is genuinely useful.
The product also turns GPT-6 Astra’s computer-use and long-horizon capabilities into a durable operating model. A dot can keep context, decide when to resume work, coordinate background agents, and return only when a result or decision needs attention.
OpenAI’s safety documentation says dots were tested against prompt injection, impersonation, and data-exfiltration attempts. The company also strengthened confirmation policies after testing exposed weaknesses around sensitive disclosures—evidence that persistent agents require operational controls beyond ordinary chat safeguards.