Key takeaways
- A personal AI account puts company data under one person's terms and settings, where the business cannot see it, audit it or delete it.
- The account type sets the rules, not the model. Personal plans, business workspaces and API access each treat training, retention and control differently.
- Bans rarely hold. A one-page policy, a sanctioned tool as good as the personal one, and API-backed tools for repeat work do.
Someone on your team has a quote to write, a long email thread to summarize or a spreadsheet to clean up. They open the AI tool they already pay for at home, paste in the material and have a good answer in a minute. Nobody is being careless on purpose. They found a faster way to do their job.
This is often called shadow AI: AI use that happens outside the tools and rules the company has approved. The trouble is not that people use AI. It is that customer names, pricing, contracts and internal numbers now sit in accounts the company does not own and cannot see.
What goes wrong with personal logins
- The terms belong to the individualA personal account is an agreement between one person and the vendor. On many consumer plans, conversations can be used to improve future models unless the user switches that off, and the switch is theirs, not yours.
- There is no recordNo admin console, no audit log, no usage report. If a customer asks where their data went, you cannot answer.
- The data leaves with the personChat history, uploaded files and saved instructions stay in the personal account when an employee moves on.
- Your obligations still applyNDAs, customer contracts and privacy rules follow the data wherever it goes. None of them has an exception for a chat window.
Data governance comes first
Data governance is the plain business of knowing what data you hold, who may use it and where it is allowed to go. Many companies have run on informal answers to those questions for years. AI makes informal answers expensive, because it is the easiest route company data has ever had out of the building: copy, paste, send.
A workable start is four classes of data, each with a rule for AI. The point is that anyone on the team can decide in a few seconds whether something may be pasted, and where.
| Class | Examples | Where it may go |
|---|---|---|
| Public | Published catalogs, marketing copy, website content. | Any tool. |
| Internal | Procedures, meeting notes, routine project details. | Company-managed AI accounts. |
| Confidential | Customer lists, pricing, contracts, financials. | Approved tools under a business agreement, or API-backed applications. |
| Restricted | Personal data, credentials, regulated or legally privileged material. | Purpose-built systems with documented controls. Never a chat window. |
Three ways to reach a model, three sets of rules
The same AI model can be reached three ways, and the route decides how your data is treated.
| Question | Personal account | Business workspace | API inside your own application |
|---|---|---|---|
| Who agrees to the terms | The individual. | The company. | The company. |
| Training on your data | Often on unless the user opts out. | Off by default. | Off by default. |
| Admin control and audit | None. | Single sign-on, user management and usage logs. | Built in: every request can be logged against a user. |
| What the model sees | Whatever gets pasted. | Whatever gets pasted or connected. | Only the fields the application sends for the task. |
| When someone leaves | The history goes with them. | The account is disabled. | There is nothing to take. |
Based on the published terms of the major model providers at the time of writing. Terms change, so confirm the agreement for the plan you are on.
API access is the route most teams overlook. Under their commercial terms, the major model providers do not train on data sent through the API by default, and they publish how long requests are retained. Some offer zero-retention arrangements.
More important is what the API lets you build. The model becomes a component inside your own software, not a destination people carry data to. The application keeps the key on the server, sends only the fields a task needs, applies the signed-in user's access rules, logs each request and puts the result in front of a person to approve. Changing models later is a settings change. We describe this pattern in our whitepaper, Custom-built software vs. mass-market SaaS.
One caution: API keys can go rogue the same way logins do. An analyst with a personal key in a script is the same problem behind a different interface. Keys should be issued by the company, kept on a server, capped by a spending limit and rotated when people change roles.
A practical plan for managing it
- Find out what is in useAsk, without penalty: which tools, which accounts, which tasks. The answers double as a list of where AI is already paying off.
- Write a one-page policyThe data classes, the approved tools and the short list of things that never go into any AI tool. Past one page, it will not be read.
- Offer a sanctioned path that is as goodPeople use personal accounts because nothing better was provided. A company workspace with single sign-on, on a plan that keeps data out of training, removes most of the reason.
- Move repeat work into API-backed toolsWhen the same task shows up every day (reading purchase orders, summarizing threads, drafting quotes), build it into the system where the work happens, with the controls included.
- Own the accounts and the keysCompany billing, company identities, and offboarding that covers AI tools along with email.
- Review on a scheduleCheck usage logs and vendor terms each quarter. Both change faster than an annual policy cycle assumes.
Govern the path, not the person
A ban on AI mostly moves it to phones and home laptops, where you see even less. The teams that handle this well do the opposite: they make the governed way the easiest way. Start with the survey and the one-page policy. Both can be done this month, and neither needs new software.
Need a governed way to put AI to work? LekkerLogic builds AI tools that reach models through business APIs, under your access rules, with data flows documented for IT and legal review.
Start a conversationThis article is general guidance, not legal advice.