Prompting ChatGPT typically requires sitting at your keyboard, giving instructions, and waiting for an output. OpenAI’s new agentic system flips that dynamic entirely by assigning AI active jobs that run continuously in the background.
OpenAI has launched Dots, autonomous AI agents powered by GPT-6 Astra that execute background tasks inside isolated cloud environments across thousands of connected software apps. Currently limited to ChatGPT Pro, Business Premium, and Enterprise subscribers, this restricted release stems from security safeguards, complex authorization boundaries, and significant computational resource requirements.
How Dots Differ From Standard ChatGPT
Standard AI chat interfaces operate on a simple request-and-response loop, pausing completely once an answer is generated. A Dot operates on a dedicated virtual machine equipped with a cloud-based web browser, allowing it to maintain ongoing progress toward user objectives even when the user closes their browser window.
Integrated directly into platforms like Slack, Microsoft Teams, and ChatGPT, Dots leverage an ecosystem of over 4,000 plugins. Rather than waiting for explicit commands, these agents perform proactive research. For instance, a Dot can identify an unbilled client, draft a proper invoice, and queue it up for human approval without being asked.
Security Guardrails and Technical Challenges
Granting an AI system persistent access to real-world applications increases operational risks. If a standard chatbot misinterprets a query, it outputs inaccurate text. If an autonomous agent misinterprets an instruction while managing connected software, it could alter records or initiate unauthorized actions.
To mitigate these risks, OpenAI implemented specific security controls:
- Read-Only Background Mode: While conducting background research, a Dot cannot modify application data, control browsers, or send messages independently.
- Granular Custom Rules: Users define which external applications a Dot can interact with and specify which tasks require mandatory human approval.
- Automated Review Systems: Sensitive operations, such as credential management or external data sharing, trigger automatic security checks against safety guidelines.
OpenAI’s internal benchmark testing highlighted the complexity of authorization safety. When evaluating the underlying GPT-6 Astra model in simulated workplace settings, researchers observed a 3% rate of misaligned actions under default security settings. While this represents a notable improvement over the 8% misaligned rate logged by GPT-5.6 Sol under identical conditions, it underscores the ongoing challenge of maintaining strict task boundaries.
Overview: ChatGPT Standard vs. OpenAI Dots
| Feature | Standard ChatGPT | OpenAI Dots |
|---|---|---|
| Operating Model | Prompt-and-response (Reactive) | 24/7 Background execution (Proactive) |
| Runtime Environment | Standard chat interface session | Dedicated cloud browser and computer |
| Supported Tiers | Free, Go, Plus, Pro, Business, Enterprise | Pro, Business Premium, Enterprise (Beta) |
| App Integration | Direct prompt tool calling | 4,000+ plugins via Slack, Teams, ChatGPT |
| Safety Architecture | Input/output content filters | Read-only background execution, custom action rules, auto-reviews |
Compute Overhead and Tiered Availability
Running a persistent agent demands vastly more processing capacity than generating single text responses. Each active Dot consumes dedicated cloud hardware, which likely explains why initial rollout privileges remain gated behind higher-tier subscription models.
Pro and Business Premium members receive one Dot allocation within their subscription plan, alongside a dedicated operational quota for deeper research tasks. Users will eventually have options to purchase additional Dots or boost their processing speeds, confirming that continuous task execution carries a measurable computational infrastructure cost.
OpenAI has indicated that access will expand to additional users over time, though no timeline has been established for Plus, Go, or Free account availability.
Current Alternatives for Automation
Users without immediate access to Dots can utilize existing workflow automation features within the platform:
- ChatGPT Work: Handles multi-step jobs across files, web tools, and data sources within a single active session.
- Scheduled Tasks: Runs specific queries or updates on recurring intervals to deliver automated summaries and briefings.
- Meta Muse: Offers an alternative perspective on agentic workflows for users exploring persistent AI assistants.
While these intermediate tools offer valuable workflow relief, they lack the full autonomous background research and native application control engineered into the Dots platform.
Frequently Asked Questions
What are OpenAI Dots?
Dots are autonomous AI agents integrated into ChatGPT that run on dedicated cloud machines. They can connect to software tools, execute multi-step workflows, and perform proactive background research without requiring constant user prompts.
Who can access OpenAI Dots right now?
Currently, Dots are available only to subscribers on ChatGPT Pro and Business Premium plans in eligible regions. Enterprise workspace members on Edu and Healthcare plans can also access the beta if an account administrator enables the feature.
Why aren’t Dots available on ChatGPT Plus or Free accounts yet?
OpenAI is conducting a phased rollout due to extensive security testing surrounding autonomous authorization boundaries and the substantial cloud computing power required to run continuous background agents 24/7.