10.3 The route matters as much as the product

In everyday life. A bank is a safe place for money. That does not make every way of sending money to it safe. You can pay through the bank's website on your own computer, or on a shared computer in a cafe. Both reach the bank. Only the first is a safe way in.

What you will learn. How to check whether the way data reaches an AI system is approved for that data, and what each AI vendor's privacy settings do and do not decide. You practice it in steps 2 and 3 of the lab.

Why it matters. Approving a product is not enough, because one product offers many ways in.

A route is the way data reaches an AI system: a chat, a project, an uploaded file, a connector, which links an AI tool to other apps, or an API call from another program. So the useful question is not "Is this AI product approved?" It is "Is this route approved for this data and this purpose?"¹ Routes differ in who can see the data and how long it is kept. They also differ in whether it may be used for training, to improve the AI vendor's models, and in what the company can check later. The same AI vendor can offer a route your policy approves and one it does not. In this chapter's story, Maria, the office manager at Brightline, uploaded the company's vendor file to her personal AI account. Her mistake was not the AI vendor. It was the personal account.

Privacy controls each answer a smaller question than permission. A temporary chat keeps data out of history. Turning training off keeps it out of training. Neither decides whether the data was allowed in. An instruction to the AI such as "Do not keep this" is a wish, not a control.¹

The two boxes below show what each AI vendor's own pages said: Anthropic, which makes Claude, and OpenAI, which makes ChatGPT. Names such as Free, Pro, Team, Business and Enterprise are their plans.

Anthropic, as verified 8 October 2026.

• Training. On Free, Pro and Max, each user's privacy setting decides whether chats may be used to improve Claude. Feedback a user sends, and chats flagged for a safety review, can be used even with it off.² With it on, chats can be kept in de-identified form for up to five years.³ On Team, Enterprise and the API, inputs and outputs are not used for training by default, unless a user sends feedback or the organization opts in.⁴
• Incognito chats. Not used for training, not saved to history, not used for memory, and kept for 30 days by default. Not available inside projects.⁵
• Retention. On individual plans, deleted chats are removed from back-end systems within 30 days, with exceptions. Chats flagged under the Usage Policy can be kept up to two years, and feedback up to five years.³ Enterprise Owners can set a custom retention period of at least 30 days. Without one, Enterprise data is kept with no time limit.⁶
• Memory. On Team and Enterprise, Owners control whether memory is available, and it is off for each member until the member turns it on.⁷
• Records. Audit logs are for Enterprise only, also through the Compliance API.⁸ On Team and Enterprise, the Primary Owner can export the organization's data.⁹
• Connectors. On Team and Enterprise, an Owner must enable a connector before members can use it. Owners can set each of a connector's tools to always allow, needs approval or blocked.¹⁰
• Usage Policy. Some uses that reach consumers are high-risk: legal, healthcare, insurance, finance, employment and housing, academic testing, accreditation and admissions, and media content generated and published automatically. In these areas, a qualified professional in that field must review advice or decisions that directly affect individuals or consumers before they are shared or made final. When outputs reach people directly, they must be told AI is involved, at least at the start of each session.¹¹

OpenAI, as verified 8 October 2026.

• Training. On plans for individuals, the "Improve the model for everyone" setting decides it. When it is off, new conversations are not used for training. On Business, Enterprise and Edu, business data is not used for training by default.¹²
• Temporary Chat. Not in history, does not create memories, not used to improve models. It may be kept for up to 30 days for safety.¹³ A chat cannot be added to a project while it is temporary.¹⁴
• Retention. Workspace admins on Business and Enterprise control how long data is kept.¹⁵
• Memory. Business workspace owners can turn memory and personalization off for the whole workspace.¹⁶ On Enterprise, admins control memory through workspace and role settings.¹⁷
• Records. Enterprise offers audit logs through the Compliance API, and an Admin Console with single sign-on and user provisioning.¹⁸
• Apps. On Business, apps are on by default and admins can change that. On Enterprise and Edu, new apps are off by default, with role-based access.¹⁹
• Usage Policies. They prohibit automating high-stakes decisions in sensitive areas, including credit, employment, housing, legal, medical, insurance and education, without human review.²⁰

What differs, and why it matters. Two differences change a decision. First, a setting for how long data is kept comes only with Enterprise at Anthropic, but already with Business at OpenAI. A company that must limit how long chats are kept needs to check the plan, not just the AI vendor. Second, defaults for new connectors, which OpenAI calls apps, differ by plan. Read the default before you assume a new connector is off. Two things are the same on both. The owners of a plan for companies can turn memory off for everyone. And neither AI vendor lets a chat inside a project be temporary or incognito. So work in a company project leaves a chat history on both, unless someone deletes it or a retention setting removes it.

What stays true. No setting on either AI vendor answers whether this data may go through this route for this purpose. Your policy and the data's owner in your company answer that. The settings carry out the answer.

Retention, audit logs and export differ by plan and change often. Check your own plan before you rely on them.