Skip to content
IOmergent

AI accounts explained

Which AI account is which?

The same product name can mean four different things. Claude, ChatGPT, Gemini and Copilot each come as a personal app, a company workspace, a developer platform and an agent on your computer. They look alike, but they differ in who controls them, what they can reach, and whether your data trains the model. Here's how to tell them apart.

In short: every vendor sells four kinds of AI product: a personal app, a company workspace, a developer platform and an agent on your computer. Business, enterprise and API plans don't train on your data; personal plans may, unless the person turns it off.

By Jon Rose, Founder & Managing Partner, IOmergent. Vendor facts as of October 2026, from the same content as our free AI Hardening Check. Spot something out of date? Tell us.

1. The four kinds

Every vendor sells the same four kinds of product. Knowing which kind you're using tells you who controls it and what it can reach.

KindExamplesWho controls itWhat it can reach
Personal appClaude Free, Pro and Max; ChatGPT Free to Pro; Gemini; Microsoft CopilotThe personTheir chats and uploads, plus any apps they connect
Company workspaceClaude Team and Enterprise; ChatGPT Business and Enterprise; Gemini in Google Workspace; Microsoft 365 CopilotYour adminsCompany chats, and the company data admins allow (connectors, files, email)
Developer platformClaude Console; OpenAI Platform; Google AI Studio and Vertex AI; Azure OpenAIYour developersWhatever your apps send it, through API keys
Agent on your computerClaude Code; Codex; Cursor; GitHub Copilot; Gemini CLI; Claude DesktopEach person's settings files, unless IT enforces themYour files, your terminal, and any tools you connect

2. Quick truths

  • A work email doesn't make a work account. The same email can have a personal Claude account and a Team account side by side, switched in the account menu. Company settings only cover the work one.
  • Personal plans may train on your chats. Unless the person turns it off. Business, enterprise and API plans don't train on your data.
  • Agents act, not just answer. They can read files, run commands and use connected tools, so their settings matter more than any chat app's.

3. Anthropic (Claude)

Personal claude.ai

Personal app

One person's Claude account.

Can reach
Their chats and uploads, memory, apps they connect, and Claude in Chrome.
Who controls it
The person.
Plans
Free, Pro ($20/mo), Max (from $100/mo); Claude Code in Pro/Max
Trains on your data
User's choice ('Help improve Claude')
How long it keeps data
5 years if training allowed, else 30 days
Company sign in
None

The AI Hardening Check covers 8 settings here

claude.ai Team

Company workspace

A company Claude workspace for 2 to 150 seats.

Can reach
Company chats and projects, and the connectors admins allow. Code execution, file creation and Skills are on by default.
Who controls it
Your admins.
Plans
Standard $25/mo ($20 annual), Premium $125 ($100 annual); 2 to 150 seats
Trains on your data
No (commercial terms)
How long it keeps data
Deleted chats purged within 30 days
Company sign in
SSO yes; SCIM no; with a verified domain, 'Restrict organization creation' blocks new personal orgs

The AI Hardening Check covers 25 settings here

claude.ai Enterprise

Company workspace

Team plus SCIM, domain capture (which retires staff's personal accounts), audit logs and custom retention.

Can reach
The same as Team.
Who controls it
Your admins, from your identity provider.
Plans
$20/seat/mo annual, 20-seat minimum, plus usage at API rates
Trains on your data
No
How long it keeps data
Custom retention (minimum 30 days); no ZDR for claude.ai chats (Claude Code on Enterprise has its own ZDR offer)
Company sign in
SSO, SCIM, domain capture

The AI Hardening Check covers 25 settings here

Claude Console (API)

Developer platform

For building apps on Claude. Separate from claude.ai, with its own members, API keys and spend limits.

Can reach
Whatever your apps send it, through API keys.
Who controls it
Your developers and Console admins.
Plans
Usage-based; tiers Start ($500/mo cap), Build ($1,000), Scale ($200,000), Custom
Trains on your data
No
How long it keeps data
API inputs/outputs deleted within 30 days; ZDR by agreement
Company sign in
SSO via a parent organization (from claude.ai Team or Enterprise); a standalone Console org asks Anthropic sales to create one

The AI Hardening Check covers 11 settings here

Claude Code

Agent on your computer

Claude's coding agent in the terminal, the IDE and the desktop app.

Can reach
Files in the projects it opens, your terminal, and the tools you connect (MCP servers, plugins, hooks).
Who controls it
Each person's settings files, unless IT deploys managed settings.

The AI Hardening Check covers 15 settings here

Claude Desktop

Agent on your computer

The Claude chat app on your computer, with extensions and Cowork.

Can reach
Local tools and MCP servers you add, folders you give Cowork, and computer use if it's turned on.
Who controls it
Each person, unless IT locks it to the company org.

The AI Hardening Check covers 6 settings here

4. OpenAI (ChatGPT)

Personal ChatGPT

Personal app

One person's ChatGPT: Free, Go, Plus or Pro.

Can reach
Their chats and uploads, memory, and apps they connect.
Who controls it
The person.
Plans
Free, Go, Plus, Pro
Trains on your data
Opt-out toggle 'Improve the model for everyone'; default on
How long it keeps data
Temporary Chat not saved or trained on
Company sign in
None

The AI Hardening Check covers 8 settings here

ChatGPT Business

Company workspace

A company ChatGPT workspace, paid per seat. Formerly ChatGPT Team.

Can reach
Company chats, the apps and connectors admins allow (company knowledge apps are on by default), custom GPTs and Codex.
Who controls it
Your admins.
Plans
Per seat, monthly or annual
Trains on your data
No
How long it keeps data
No custom retention: chats are kept until the user deletes them
Company sign in
SSO (SAML/OIDC) and domain verification; SCIM no

The AI Hardening Check covers 17 settings here

ChatGPT Enterprise and Edu

Company workspace

Business plus SCIM, custom roles, custom retention, IP allowlisting and compliance logs.

Can reach
The same as Business, but company knowledge apps must be turned on by an admin.
Who controls it
Your admins, from your identity provider.
Plans
Quote only
Trains on your data
No
How long it keeps data
Custom retention
Company sign in
SSO, SCIM, domain verification

The AI Hardening Check covers 17 settings here

OpenAI Platform (API)

Developer platform

For building apps on OpenAI models. Billed separately from ChatGPT.

Can reach
Whatever your apps send it, through API keys.
Who controls it
Your developers and organization owners.
Plans
Usage-based, with usage tiers that raise rate limits
Trains on your data
No
How long it keeps data
Abuse-monitoring logs up to 30 days; ZDR / modified abuse monitoring per org or project with approval
Company sign in
SSO with a verified domain (Enterprise or custom billing plan)

The AI Hardening Check covers 9 settings here

Codex

Agent on your computer

OpenAI's coding agent: CLI, IDE extension, desktop app and cloud tasks.

Can reach
Files in the projects it opens, your terminal, and connected tools.
Who controls it
Each person's config file, unless IT deploys managed requirements.

The AI Hardening Check covers 6 settings here

ChatGPT desktop app

Agent on your computer

ChatGPT on your computer.

Can reach
Apps it works with on your screen, and computer use if it's allowed.
Who controls it
The person, or IT through a managed policy.

The AI Hardening Check covers 3 settings here

5. Google (Gemini)

Gemini with a personal Google account

Personal app

The Gemini app signed in with a personal Google account.

Can reach
Chats, connected Google apps (Gmail, Drive, Photos, YouTube), personal context and memory, and Gems.
Who controls it
The person.

The AI Hardening Check covers 7 settings here

Gemini in Google Workspace

Company workspace

The Gemini app and Gemini in Gmail, Docs and Drive for your company's domain.

Can reach
Company Workspace data the person can already open, and the third party integrations admins allow.
Who controls it
Your Workspace admins, per organizational unit or group.
Plans
Gemini app is a core service on Business Starter/Standard/Plus, Enterprise, Essentials, Frontline and Nonprofits; on other editions it is an additional service under consumer terms
Trains on your data
No training on customer data without permission; Gemini app chats not human-reviewed
How long it keeps data
Gemini app 3, 18 or 36 months (default 18); Workspace apps 90 days to indefinite; up to 72 hours with history off
Company sign in
SSO via Security > Authentication > SSO with third party IdP, per OU or group

The AI Hardening Check covers 27 settings here

Google AI Studio and Vertex AI

Developer platform

For building with Gemini models: AI Studio for quick API keys, Vertex AI on Google Cloud.

Can reach
Whatever your apps send it. Use AI Studio's paid tier for work data.
Who controls it
Your developers and Google Cloud admins.

The AI Hardening Check covers 8 settings here

Gemini CLI and Code Assist

Agent on your computer

Google's coding agent in the terminal and the IDE.

Can reach
Files in the projects it opens, your terminal, and connected tools.
Who controls it
Each person's settings, unless IT deploys a system settings file.

The AI Hardening Check covers 7 settings here

6. Microsoft (Copilot)

Microsoft Copilot (personal)

Personal app

Copilot with a personal Microsoft account, including in home Office apps.

Can reach
Chats, memory, and Gmail, Drive and Outlook if the person connects them.
Who controls it
The person.

The AI Hardening Check covers 9 settings here

Microsoft 365 Copilot

Company workspace

Copilot in your company's Microsoft 365: Copilot Chat for everyone, and licensed Copilot in Word, Outlook and Teams.

Can reach
Everything the person can already open in Microsoft 365, so oversharing in SharePoint and OneDrive shows up in answers.
Who controls it
Your Microsoft 365 admins.
Plans
Microsoft 365 Copilot is included in E7 and an add-on license elsewhere; Copilot Chat agents can bill pay-as-you-go
Trains on your data
Prompts, responses and Microsoft Graph data aren't used to train foundation models
How long it keeps data
Set with Purview retention policies for Copilot interactions
Company sign in
Microsoft Entra ID; MFA and Conditional Access recommended

The AI Hardening Check covers 24 settings here

Copilot Studio

Developer platform

Where your staff build Copilot agents.

Can reach
The connectors and knowledge sources each agent is given.
Who controls it
Your Power Platform admins.

The AI Hardening Check covers 8 settings here

Azure OpenAI and Foundry

Developer platform

OpenAI and other models inside your Azure subscription.

Can reach
Whatever your apps send it, within your Azure network settings.
Who controls it
Your Azure admins.

The AI Hardening Check covers 9 settings here

7. GitHub Copilot

Copilot Business and Enterprise

Company workspace

Copilot managed by your GitHub organization.

Can reach
Your repositories (minus any content exclusions), the MCP servers your policy allows, and the cloud agent.
Who controls it
Your organization and enterprise owners.

The AI Hardening Check covers 20 settings here

Copilot in VS Code and Copilot CLI

Agent on your computer

Copilot's agent on your computer.

Can reach
Files in the projects it opens, your terminal, and connected tools.
Who controls it
Each person's settings; Enterprise can enforce managed settings.

8. Cursor

Cursor Pro

Agent on your computer

The Cursor editor and agent on a personal plan.

Can reach
Files in the projects it opens, your terminal, and connected tools.
Who controls it
The person, including Privacy Mode.

The AI Hardening Check covers 3 settings here

Cursor Teams and Enterprise

Agent on your computer

Cursor managed by your company.

Can reach
The same, plus cloud agents.
Who controls it
Your Cursor admins: Privacy Mode, allowed MCP servers and models, spend limits.

The AI Hardening Check covers 17 settings here

Common questions

Does ChatGPT train on my data?
Personal ChatGPT plans do by default, unless the person turns off 'Improve the model for everyone'. ChatGPT Business, Enterprise and the OpenAI API platform don't.
Does Claude train on my data?
On personal claude.ai plans it's each person's choice, under 'Help improve Claude'. Claude Team, Enterprise and the Claude Console (API) don't, under Anthropic's commercial terms.
Is a work email the same as a work account?
No. The same email can have a personal account and a company account side by side, switched in the account menu. Company settings only cover the company account.
What's the difference between ChatGPT Plus and ChatGPT Business?
ChatGPT Plus is a personal plan the person controls. ChatGPT Business is a company workspace your admins control, with company sign in and no training on your data.

Not sure which ones your team uses?

The free AI Hardening Check finds the AI agents on your machine and walks your admin settings, in your own AI assistant. Or talk to a CISO about the full picture.

Run the free check