OpenAI launches Dots: always-on GPT-6 Astra agents with their own cloud computers

At DevDay, OpenAI introduced Dots, always-on agents powered by GPT-6 Astra that each run on a dedicated cloud computer and reach 4,000+ apps, rolling out to Pro and business plans with safeguards including auto-review and a safety monitor.

At DevDay on September 29, 2026, OpenAI introduced dots, always-on agents powered by GPT-6 Astra that pursue user-defined goals continuously with minimal oversight, unlike a Codex or ChatGPT session. Each dot runs on its own cloud computer with a browser and reaches more than 4,000 apps via ChatGPT plugins. Users can message or voice-call a dot from ChatGPT, connect it to Slack and Microsoft Teams, and start one from Codex or ChatGPT, with SMS planned; The Verge reports only one dot can be created for now. Availability begins with ChatGPT Pro and Business Premium in eligible markets, though OpenAI's announcement also lists Enterprise, which The Verge and TechCrunch do not. Safeguards include read-only background research, custom rules, auto-review of sensitive actions such as password changes, and a safety monitor that can pause or stop a dot's work. Specialist enterprise dots, previewed for access management and deep system-of-record integrations, can receive identities, credentials, and tools, and OpenAI is working with Microsoft on Agent 365 security controls. MarkTechPost reports GPT-6 Astra scores 72.6% on OSWorld 2.0 at roughly 40 minutes per task, versus 65.7% for GPT-5.6 Sol. The Verge frames dots as a competitor to Meta's Muse, launched weeks earlier and later the subject of a safety complaint after a user said it shared their address; MarkTechPost separately notes the launch follows an agent incident that posted 53 ChatGPT users' images online.

• Dots are always-on agents powered by GPT-6 Astra, each running on a dedicated cloud computer with its own browser
• Dots reach 4,000+ apps via ChatGPT plugins
• Users can message or voice-call a dot from ChatGPT and connect it to Slack and Microsoft Teams, with SMS planned; dots can also be started from Codex
• Only one dot can be created for now, according to The Verge
• Rollout starts with ChatGPT Pro and Business Premium in eligible markets; OpenAI's announcement also lists Enterprise, which The Verge and TechCrunch do not
• Safeguards: read-only background research, custom rules, auto-review of sensitive actions such as password changes, and a safety monitor that can pause or stop a dot
• Specialist enterprise dots, previewed for access management and system-of-record integrations, can receive identities, credentials, and tools; Microsoft Agent 365 security integration is underway
• GPT-6 Astra scores 72.6% on OSWorld 2.0 at roughly 40 minutes per task, versus 65.7% for GPT-5.6 Sol