Is dots safe? Permissions, approvals and your data
Dots give you more control dials than most agents, which makes them fairly safe if you use the dials. Every action type can be set to ask first or hand off, password changes always stay with you, and your own computer is off limits unless you open it up.
Checked October 20, 2026. We summarise what OpenAI says and flag what it doesn't say. We don't speculate.
The consent form
Five things to know before you connect your accounts. Each line is from OpenAI's own pages unless it says otherwise.
- Where it runs
- Each dot has its own cloud computer and browser that you can watch or take over. Access to your own computer is off by default and needs the ChatGPT desktop app.
- What it can reach
- You choose which plugins (apps) it connects to; read access may not include permission to change data or send messages. Proactive research uses read-only connections.
- When it stops to ask you
- Custom Rules let you set each kind of action to act freely, act only if you pre-approved it, ask first, or hand off to you. An auto-review check screens actions that touch accounts or share information, and some steps like password changes always stay with you.
- What happens to your data
- Business, Enterprise and Edu content is not used for training by default; personal-plan users can control training use. OpenAI says it does not train directly on proactive research or a dot's own notes. No retention period is published.
- How to cut it off
- Disconnect apps in ChatGPT's app controls (this does not delete data already gathered), revoke local computer access in the Computers section, pause the dot, or Reset it, which deletes its conversations, memories and scheduled tasks.
How to use it carefully
Each dot works on its own OpenAI cloud computer and browser, which keeps its sign-ins separate from your personal browser. You pick the plugins it connects to, and read access does not by itself let it change data or send messages. Proactive research uses read-only connections. Custom Rules set each action to act freely, act if pre-approved, ask first or hand off, and an auto-review step checks actions that affect accounts or share information. Business, Enterprise and Edu content is not used for training by default, and personal-plan users can control it. OpenAI publishes no retention period. To pull back, disconnect apps in ChatGPT's app controls, revoke laptop access under Computers, pause the dot, or Reset it. Only a Reset deletes its conversations, memories and scheduled tasks.
Sources
- Custom Rules let users set actions to proceed, proceed only if pre-approved, ask first, or hand off, but cannot override built-in safeguards such as leaving password changes to the user. help.openai.com official, read Oct 20, 2026
- An auto-review step checks actions that could affect accounts or share information before they happen. openai.com official, read Oct 20, 2026
- Disconnecting an app does not erase what the dot already learned from it; only a Reset, which deletes the dot's conversations, memories and schedules, does that. help.openai.com official, read Oct 20, 2026
- Local computer access is off by default and can be revoked from the Computers section; the cloud browser keeps its own sign-ins separate from yours. help.openai.com official, read Oct 20, 2026
- Workspace content from Business, Enterprise and Edu is not used for training by default, and personal-plan users can control training use; no retention period is stated. openai.com official, read Oct 20, 2026
- OpenAI's allowlisting guide says ChatGPT's cloud browser signs requests under RFC 9421 with a Signature-Agent value of "https://chatgpt.com", with keys at chatgpt.com/.well-known/http-message-signatures-directory; the page does not name dots specifically. help.openai.com official, read Oct 20, 2026