Knowing what information employees should not paste into public AI chatbots is an important part of using these tools responsibly at work. A chatbot may be useful for drafting or brainstorming, but a prompt can also contain details your business has a duty or strong reason to protect.
A practical default is simple: use a public chatbot only with information approved for public release, unless your organization has specifically approved the tool and the data use. That rule gives employees a clear starting point while your business evaluates how different AI tools handle prompts, files, retention, and access.
Information employees should keep out of public AI chatbots
Personal identifiers and private records
Do not paste customer, patient, employee, student, or other individual-level information into a public chatbot. This includes names, contact details, account numbers, identification numbers, personnel files, private correspondence, and records about someone’s health, finances, education, or work.
Removing a name is not always enough. A distinctive job title, a small location, a rare event, or a combination of details may still make someone recognizable. Attachments can also contain identifiers in the text, file name, comments, or document properties. If a task depends on real personal information, use an approved process rather than assuming that a quick edit makes it safe.
Regulated or contract-protected data
Keep information subject to privacy obligations, contracts, or sector-specific requirements out of a public chatbot unless the tool and use have been reviewed and approved for that purpose. This may include protected health information, education records, payment-related information, or records covered by a customer agreement. The rules that apply depend on the organization, the data, and the circumstances.
For example, the U.S. Department of Education describes privacy and data-sharing considerations for education records in its guidance on privacy and data sharing. Treat the sensitivity of a record as a reason to check your organization’s approved process, not as a prompt to test what a chatbot will accept.
Passwords, keys, and other secrets
Never enter passwords, passphrases, multifactor authentication codes, API keys, private certificates, recovery codes, or session tokens. These are credentials or secrets that can provide access to business systems. The same caution applies to configuration files, scripts, and screenshots that may contain them.
If an employee needs help troubleshooting, they can describe the issue without including the secret itself. Replace sensitive values with clearly fictional placeholders, and check that the surrounding material does not reveal a working account, system, or connection.
Confidential business material
Do not paste information that is not meant for public release, such as internal financials, customer lists, pricing plans, proposals, contract terms, unpublished product plans, acquisition or hiring discussions, legal work product, or proprietary processes. Source code and internal technical documentation may also reveal intellectual property or how a business operates.
When in doubt, ask whether the information would be appropriate to publish on the company website or share with an unknown person outside the organization. If not, do not put it in a public chatbot without explicit approval.
Security and operational details
Keep network diagrams, firewall rules, system configurations, vulnerability reports, incident details, backup procedures, and access-control information out of public tools. Individually, a detail may seem harmless. Taken together, operational information can reveal how business systems are organized or where they may be exposed.
Payment environments deserve particular care. The PCI Security Standards Council discusses considerations for securing AI use in payment environments in its AI principles article. Follow your organization’s approved handling procedures for payment-related information rather than putting it into an unreviewed tool.
Public chatbots and approved business AI are not the same
“AI chatbot” does not describe one uniform set of privacy and security practices. A public service and an organization-approved system can differ in their settings, contract terms, retention practices, access controls, and permitted uses. Do not assume that a tool protects business information just because it is popular, has a private mode, or is used through a work account.
Before approving an AI tool, the people responsible for technology, security, privacy, and relevant business operations should understand what data can be entered, how prompts and uploaded files are handled, who can access them, and whether the tool is suitable for the intended work. The NIST Generative AI Profile offers a risk-management reference for organizations evaluating generative AI.
Safer ways to get useful results
- Use a synthetic example. Ask for help using fictional names, sample figures, and made-up scenarios instead of real records.
- Remove unnecessary details. Share only what the task requires, and check for identifying combinations, hidden document details, or embedded credentials.
- Ask for a generic template. Request a general checklist, outline, or draft structure, then complete it inside an approved business system.
- Use an approved tool for approved data. Confirm the specific tool and intended data use are authorized, rather than treating approval of one AI product as approval of every product.
- Pause when uncertain. Ask a manager or the person responsible for security or privacy before submitting information you would not share publicly.
Make the employee policy specific and usable
An employee AI acceptable use policy should do more than say “be careful.” It should identify which tools are approved, which information types are prohibited, and who can authorize exceptions. Give employees examples that match their work, including customer records, employee files, credentials, contracts, source code, and security configurations.
Explain how to sanitize a task, where employees can get help, and what to do if they make a mistake. Keep the guidance easy to find and reinforce it during onboarding and when approved tools or business practices change. For broader security planning, Zevonix’s IT risk management overview can help frame how technology risks fit into business decisions.
What to do after an accidental disclosure
Tell your manager or designated security contact promptly. Share what was entered, which tool was used, when it happened, and whether a file was attached. Do not repeat the sensitive material in follow-up messages or try to conceal the incident.
Follow your organization’s incident process. The appropriate next steps depend on the information and tool involved, so avoid making assumptions about whether the data was retained, accessed, or used. The business can then assess the situation, document it, and decide whether further action is needed.
Frequently asked questions
Is it safe to paste customer or employee personal information into an AI chatbot?
Do not put personal information into a public chatbot unless your organization has approved that specific tool and use. Removing names alone may not prevent someone from being identified.
Can employees enter patient information or other protected health information into public AI?
Keep it out of public tools unless your organization has specifically reviewed and approved the tool and workflow for that information. Ask your privacy or compliance contact when unsure.
Can a public AI chatbot use prompts or uploaded files to improve its models?
Practices and settings can differ by tool. Review the specific service’s terms and your organization’s approval before sharing business information; do not assume every chatbot handles prompts the same way.
What information can employees use if they remove names and identifiers?
Only use information that is approved for the tool and task. A combination of details, a unique case, or information left in an attachment may still identify a person or expose confidential facts.
What should an employee do after accidentally sharing confidential information?
Notify the designated manager or security contact promptly and provide the details needed to assess the incident. Follow your organization’s response process.
Keep AI use useful and controlled
Clear boundaries let employees benefit from AI tools without treating sensitive information casually. Start with a public-release rule, define approved tools and data uses, and make the reporting process straightforward. If your business needs help aligning technology practices with its security and operational goals, contact Zevonix at (904) 658-0777.