Public AI tools can help employees brainstorm, summarize, and draft content. But a prompt can also expose information that should stay within your business. If you are deciding what not to enter into public AI tools, start with a simple rule: do not share information that could identify a person, provide access to an account, or reveal something confidential unless the tool has been approved for that data and use.
“Public” usually means a consumer website or app that employees access on their own, rather than a tool selected and governed by the business. AI tools do not all handle prompts in the same way. Their settings, account types, data retention practices, and terms can differ, so employees should not assume that a prompt is private just because it was entered for work.
Information employees should keep out of unapproved AI tools
Make the following a practical red list for public or unapproved tools:
- Passwords and access credentials: Keep passwords, authentication codes, API keys, private keys, recovery codes, and login links out of prompts. These details can provide access to business systems or data.
- Identifiable customer or client information: Do not paste names, email addresses, phone numbers, account details, support messages, case notes, or records that identify a customer or client.
- Patient, student, or employee records: Keep health information, education records, personnel files, performance notes, payroll details, and other records about individuals out of unapproved tools.
- Payment and financial information: Do not submit bank or payment account details, transaction records, financial statements, or sensitive billing information.
- Confidential business documents: Avoid uploading contracts, legal material, proposals, internal reports, pricing information, meeting notes, or documents marked confidential.
- Trade secrets and technical details: Protect unreleased plans, proprietary processes, source code, product designs, security configurations, and information about systems that could help someone gain access.
These categories are a starting point, not a complete inventory. NIST’s Generative AI Profile includes data privacy among the risks organizations should consider when managing generative AI.
Removing names may not be enough
Employees sometimes try to make a prompt safer by deleting names. That can help in some situations, but it does not automatically make the information anonymous. A combination of details, such as a job title, location, unusual event, date, or specific customer issue, may still point to a particular person or organization.
For example, a request that includes a customer’s exact complaint, purchase details, and a distinctive business situation could reveal who the customer is even without a name. The same concern applies to employee examples and patient, student, or client information. When in doubt, treat details that could identify someone as sensitive.
Can employees paste customer emails or upload documents?
Employees should not paste customer emails, records, or confidential documents into a public AI tool unless the business has specifically approved that tool and the particular use. A helpful rewrite or summary is not worth exposing information the organization is responsible for protecting.
The same principle applies to business documents. A public tool may be convenient for summarizing a contract or finding a phrase in a report, but the document could include private customer details, internal plans, or sensitive terms. Employees should check the organization’s approved tools and rules before uploading any file or copying text into a prompt.
Approval should be based on more than whether a tool is popular or has a business-looking interface. The business should review the provider’s terms and settings, including how prompts are retained or used, who can access submitted information, and what protections apply to the account. A paid or employer-provided account is not automatically approved for every kind of data.
Safer ways to get useful AI assistance
When employees need help with a task, they can often describe the problem without sharing the source material:
- Use a fabricated example. Replace real people, organizations, dates, and records with made-up details that preserve the structure of the task.
- Ask a general question. Request a template, outline, checklist, or explanation without including internal facts or customer information.
- Summarize at a high level. Share only the broad issue when the task does not require exact details.
- Remove identifying details carefully. Check for indirect identifiers and combinations of details, not just names and email addresses.
- Use an approved tool for approved work. Follow your organization’s rules about which accounts and data types are permitted.
For example, instead of pasting a real customer complaint, an employee could ask for a neutral response template for a delayed service issue. A fictional scenario can provide enough context to improve the draft without exposing the customer’s message.
For more context on adopting AI with care, see Zevonix’s article on starting AI automation safely. Businesses reviewing their broader safeguards may also find the guidance on cybersecurity for businesses useful.
Make the rules easy to follow
A business AI acceptable use policy should be short enough for employees to understand and specific enough to guide real decisions. It can explain:
- Which AI tools employees may use for work.
- Which information must never be entered into public or unapproved tools.
- What data, if any, may be used with approved tools, and for which tasks.
- How employees should check a tool’s account and data-handling settings.
- Who can approve a new tool or a use case involving sensitive information.
- How to report an accidental disclosure and where to get help.
Keep the policy aligned with how employees actually work. If a task cannot be completed safely in a public tool, explain the approved alternative or who can help. Clear examples make the rules more useful than a general instruction to “be careful.”
What to do if sensitive information is shared
If an employee realizes they entered sensitive information into an unapproved AI tool, the first step is to stop sharing additional material. They should promptly notify the manager or security contact named in the organization’s process and explain what type of information was shared, which tool and account were used, and approximately when it happened. They should not copy the information into another message while reporting it.
The organization can then follow its response process and decide what steps are appropriate. Employees should not try to handle the situation quietly or assume that deleting a prompt resolves it. A prompt may be subject to the tool’s settings and terms, so the business should assess the incident using the information available.
Frequently asked questions
Can public AI tools use prompts to train or improve their models?
That depends on the tool, account, settings, and provider terms. Employees should not assume prompts are excluded from retention or model improvement. Check the specific service’s current terms and approved business settings before using it for work.
Should employees enter passwords or API keys into an AI tool?
No. Keep passwords, access keys, authentication codes, and other credentials out of public and unapproved tools. Use your organization’s approved process for troubleshooting access or code.
Does removing names make customer information safe to use?
Not necessarily. Other details, or a combination of details, may still identify a person or organization. Use a fabricated example or ask a general question instead of submitting real customer information.
What if an employee accidentally shares sensitive data?
They should stop submitting information and promptly notify the designated manager or security contact. Follow the organization’s response process and provide enough information to assess what was shared without repeating the sensitive content.
Keep AI useful without losing control of business data
Employees can get value from AI while protecting sensitive information when the boundaries are clear: keep credentials, identifiable records, financial details, confidential documents, and proprietary material out of unapproved tools. Establish approved options, offer safe alternatives, and make reporting mistakes straightforward. If your business wants help reviewing its wider IT and security practices, contact Zevonix to discuss next steps.