The honest way to think about a work AI account is the way you already think about work email. Nobody is reading it over your shoulder, and the business can get to it if it has to.
What Claude lets an owner see
On Claude’s Team and Enterprise plans, the organisation owns the account and its data. Anthropic’s help pages say organisation data exports are available only to the Primary Owner, and that the export includes conversation data. An ordinary member cannot run that export, and nor can another admin.
So on a Team plan there is one person who can pull the organisation’s conversations out, by a deliberate act. That is different from a manager browsing your chats on a Tuesday afternoon.
What ChatGPT lets an owner see
OpenAI’s help centre says that on ChatGPT Business, owners and admins cannot see all private member chats by default. On Enterprise, eligible workspaces get a Compliance Platform that gives the owner, or a tool they authorise, a record of conversations for audit and legal purposes.
OpenAI’s own pages have not always described the Business plan the same way, so do not build your privacy on the fine print. The safe reading is the same as for Claude: the business owns the workspace.
What this means for you
- Use the work account for work. That is what it is for, and it is the account your employer has made safe to use.
- Keep personal matters out of it. Health questions, a job search, a dispute with a colleague: use your own account on your own device.
- Ask what the rule is. A good AI policy says who can access conversations, and when.
What I tell owners
Say it plainly before anyone starts: what you can see, what you will not look at, and what would change that. One line I keep fixed in a team rollout: nothing the AI records about how someone works is used to assess their performance. A promise like that is only as good as the person making it, so it should come from the owner, out loud.