AI is becoming a normal business tool. Staff are using it to rewrite emails, summarise documents, analyse information and speed up everyday work.
That can be useful. But there is an important distinction many businesses are missing:
Paying for an AI subscription does not automatically make it a business AI account.
A paid personal account is still a personal account
A staff member might pay for ChatGPT Plus or Pro and reasonably assume that because money is changing hands, the account is now suitable for business information.
That is not how the products are separated.
OpenAI treats ChatGPT Free, Plus and Pro used in a personal workspace differently from managed products such as ChatGPT Business and Enterprise. On personal ChatGPT workspaces, users can control whether conversations are used to improve OpenAI’s models. OpenAI says content from its business products is not used to train models by default. You can read OpenAI’s current business data commitments and data controls guidance.
So seeing ChatGPT Plus on a company credit card does not mean the business has implemented ChatGPT Business.
Claude and Gemini make similar distinctions
Anthropic also separates consumer Claude products from its commercial offerings. Claude Free, Pro and Max are consumer products, while Claude for Work and API use fall under commercial terms. Anthropic says commercial inputs and outputs are not used for model training by default unless the customer chooses otherwise or submits certain feedback. Its consumer products have their own privacy and model-improvement controls. Anthropic explains this in its model training guidance.
Google likewise applies different protections when Gemini is used through a managed Google Workspace environment. Google says Workspace customer data is not used to train generative AI models outside the organisation’s domain without permission. Google documents those protections in its Workspace AI privacy information.
The bigger problem is who controls the account
With a personal account, the employee usually controls the account and its settings.
That creates some simple but important questions:
Who created the account?
Who controls the privacy settings?
Who checks whether model-improvement settings have changed?
Who removes access when the employee leaves?
Who knows whether the employee is using their personal workspace or the company’s managed workspace?
Two employees can appear to be using exactly the same AI product while operating under different account types, different settings and different data-handling rules.
This matters because people paste real business information into AI
People do not normally set out to disclose company information. They are trying to get work done.
They paste in an email containing a customer’s name and phone number. They upload a contract. They ask AI to rewrite a support ticket. They provide a spreadsheet with customer information. They paste an internal incident report because they want a cleaner summary.
The moment that happens, the account type and the provider’s data rules stop being an abstract privacy setting.
They become part of how the business is handling information it may have an obligation to protect.
A paid account is not the same thing as a confidentiality agreement
A useful way to think about an unmanaged AI service is like a new contractor or business partner that the company has not yet vetted and that has not signed a confidentiality or non-disclosure agreement.
You probably would not email that person your full customer list, an HR file or a confidential incident report just because they promised to help you work faster.
The same thought process should apply to AI. Before providing business information, understand who operates the service, what terms apply to the account and what happens to the information after it is submitted.
Free AI does not mean nobody else can see what you type
There is another assumption worth clearing up. People sometimes picture AI training as a completely automated process where their prompt disappears into a huge bucket of data and only computers ever touch it.
That is not always the case.
OpenAI says conversations may be reviewed as part of improving its systems and enforcing safety requirements. Google says some Gemini conversations are reviewed by human reviewers, and that reviewed chats and related data can be retained for up to three years. Anthropic says consumer Claude conversations flagged for safety review may also be analysed and used to improve its safeguards.
You can read the current provider guidance from OpenAI, Google and Anthropic.
This does not mean every AI conversation is manually read by a person, and it does not mean your prompt automatically becomes publicly searchable on the internet.
It does mean that if you put personal information into a consumer AI account, particularly one where model improvement is enabled, you should stop thinking of that information as something that remains exclusively inside your business.
A safer assumption is that the information has been disclosed outside your controlled business environment and may be handled according to the provider’s retention, review and model-improvement processes.
Turning model improvement off later is also not the same thing as winding the clock back. Providers generally describe opt-out controls as affecting future use of new conversations, while information already used in training or review processes may not simply be pulled back out.
A personal AI account can also be connected to your business data
The risk is not limited to information somebody manually copies and pastes into a chat.
Depending on the product, plan, permissions and administrator settings, personal AI accounts can also connect to services such as email, calendars and cloud storage.
For example, ChatGPT can connect to services including Gmail, Google Calendar, Google Drive and Outlook where those apps are available and authorised. Google also allows personal Gemini users to connect services such as Gmail, Drive and Calendar.
That means an employee using a personal AI account may be able to authorise it to retrieve information from a work email account or cloud storage service that they already have access to.
They might ask it to summarise recent emails, find a customer document, review a spreadsheet or look through files without ever manually uploading those records one by one.
This does not bypass the permissions of the connected email or file account, and business administrators can often restrict or block these connections. But if the employee is allowed to connect the service, the amount of company information available to the AI can be much larger than a single prompt.
That is another reason the business should know which AI accounts are being used and which company systems have been connected to them. OpenAI documents how its connected apps work, while Google documents Connected Apps in Gemini for personal accounts.
Business-managed AI is about ownership and control
The reason to use a managed business AI environment is not simply to unlock another feature.
It is to give the organisation control over the accounts being used for work, the users who have access, the settings applied to those users and what happens when somebody joins or leaves.
Before a business starts writing a large AI policy, a useful first step is much simpler:
Work out what AI accounts your staff are actually using.
And do not assume that because they are paid accounts, they are business accounts.
The account is only the start
Even if the business has chosen an appropriate managed AI platform, there is another question to ask: what other apps and services are connected to it?
That is where the data path can become much harder to see. We will cover that next in Where Does Your Data Go When You Give It to an AI App?
Q10 can help businesses review their current AI use, move staff toward appropriate managed business accounts and put practical controls around how AI is used alongside existing cybersecurity and managed IT practices.
AI product terms and privacy settings change regularly. This article reflects publicly available provider information as at October 2026 and is general information rather than legal advice.