ChatGPT security settings
ChatGPT Security Settings: Top 10 for Admins
ChatGPT comes as a personal app, a company workspace (Business or Enterprise), the OpenAI API platform and agents on your computer (Codex and the desktop app). These are the 10 settings that matter most, where to find each one and what good looks like.
In short: start with (1) SSO, (2) Verified domain and (3) Owners, admins and who can invite, then work down the list below.
By Jon Rose, Founder & Managing Partner, IOmergent. Vendor facts as of October 2026. Not sure which version you have? See which AI account is which.
The top 10 ChatGPT security settings
Most important first. Our free AI Hardening Check walks you through these and the other 36, with direct links, inside your own AI assistant.
-
1SSO
- Where to find it
- Workspace settings > Identity & access > Identity & provisioning > Single Sign-On (needs a verified domain first; if read-only, a global admin manages it in the Admin Console)
- Good looks like
- SSO configured and set to required for the verified domain
- Why it matters
- One identity to disable when someone leaves
-
2Verified domain
- Where to find it
- Workspace settings > Identity & access > Identity & provisioning > Verified Domains
- Good looks like
- Company domain verified (ChatGPT then prompts work-email personal users to migrate)
- Why it matters
- Pulls work-email personal accounts into the workspace
-
3Owners, admins and who can invite
- Where to find it
- Workspace settings > Members (pending invites too)
- Good looks like
- Two named owners at most; no leavers; pending invites reviewed (Business has no switch to stop members inviting)
- Why it matters
- On Business, members can invite other members
-
4Apps and connectors, and their actions
- Where to find it
- Workspace settings > Apps (Enabled / Available) > each app > User access and Action control; Permissions & roles > Connected data
- Good looks like
- Only approved apps; actions read-only unless needed
- Why it matters
- Company-knowledge apps are on by default on Business
-
5Shared links and chat retention
- Where to find it
- No admin switch on Business (Enterprise: admin can disable sharing and set retention, 90-day minimum); users manage links in Settings > Data controls > Shared links
- Good looks like
- Business: a policy telling users not to share chats; Enterprise: sharing off, retention set
- Why it matters
- Shared links expose the whole conversation
-
6Who can create and share GPTs
- Where to find it
- Workspace settings > GPTs: creation, sharing scope, third-party GPTs, connected actions and allowed action domains
- Good looks like
- Enterprise: sharing limited to the workspace, action domains allow-listed. Business: third-party GPTs limited and a sharing policy (members can share GPTs without admin approval)
- Why it matters
- GPTs can be shared to anyone with the link or the GPT Store
-
7Local computer access, cloud browser and Computer History
- Where to find it
- Permissions & roles > Workspace capabilities: 'Allow local computer access', 'Cloud browser use', 'Cloud network access' and 'Computer History' (off by default). Per role on Enterprise/Edu; Business defaults unconfirmed
- Good looks like
- Off by default; on only for named roles that need it (Business: a deliberate workspace-wide choice)
- Why it matters
- Lets ChatGPT act on the user's machine and the open web
-
8Codex approval policy
- Where to find it
- config.toml: approval_policy (default on-request); requirements.toml: allowed_approval_policies; per project trust_level
- Good looks like
- on-request, never 'never'. Mark risky repos trust_level = "untrusted" per project; the old 'untrusted' approval_policy is retired and can stop the client starting
- Why it matters
- Keeps a person approving risky actions
-
9API key governance
- Where to find it
- Settings > Organization > General > API key governance: key creation limited to service-account keys, maximum key lifetime (both added September 2026; existing keys are not changed)
- Good looks like
- Service-account keys only; maximum lifetime set
- Why it matters
- User keys die with the user; long-lived keys leak
-
10Training on your chats
- Where to find it
- Settings > Data controls > 'Improve the model for everyone'
- Good looks like
- Off
- Why it matters
- On by default on personal plans, so work content in chats can be used for training
ChatGPT security settings: common questions
- What are the most important ChatGPT security settings?
- Start with (1) SSO, (2) Verified domain and (3) Owners, admins and who can invite. All 10 are listed on this page, most important first, with where to find each one and what good looks like.
- Does ChatGPT Business train on our data?
- No. OpenAI doesn't train on ChatGPT Business or Enterprise data. Personal plans train by default unless the person turns off 'Improve the model for everyone'.
- How long does ChatGPT Business keep chats?
- Until the user deletes them. Business has no custom retention setting; Enterprise lets admins set one.
- Does ChatGPT Business support SSO and SCIM?
- Business supports SSO (SAML or OIDC) once you verify your domain, but not SCIM. Enterprise adds SCIM.
- How can I check all the ChatGPT settings quickly?
- Add the free IOmergent AI Hardening Check to your AI assistant and ask: "Run the IOmergent AI Hardening Check for our ChatGPT workspace." It walks you through these and the other 36, with direct links, and nothing leaves your machine without your OK.
Want an expert review?
A self-check can't see which AI apps hold OAuth access to your email, files and code, or confirm the settings match what was reported. Our AI Security Assessment does both, with read-only connections and a live review.
See the AI Security Assessment