Knowing what information employees should keep out of public AI chatbots is an important part of using these tools responsibly at work. A prompt can contain more than the question itself: pasted email threads, uploaded files, screenshots, code, and background details may reveal confidential information.
A useful starting point is simple: do not put nonpublic business information into a public chatbot unless your organization has approved the tool and the specific use. Approval of a tool does not automatically make every kind of data appropriate to share. Your organization’s policies, contracts, professional duties, and applicable requirements still matter.
The NIST Generative Artificial Intelligence Profile includes privacy and security among the considerations organizations should address when managing AI risks. For employees, that means pausing to check what a prompt may disclose before submitting it.
Information to keep out of public AI chatbots
Use this checklist when writing a prompt, attaching a document, or copying material from another system. When in doubt, leave the information out and ask the appropriate person in your organization.
Customer and patient information
Keep customer and patient records out of public tools. This includes names and contact details, account or case information, health details, appointment records, messages, and documents that connect a person to a service or issue. Healthcare teams should follow their organization’s policies and applicable requirements, including any relevant privacy and security obligations.
Removing a name does not always make information safe. A combination of location, job role, dates, unusual circumstances, or details from a conversation may still point to a specific person. Avoid sharing the underlying record, even when direct identifiers have been removed.
Employee and applicant information
Do not paste personnel files, resumes, interview notes, performance reviews, compensation details, leave information, workplace complaints, or other private employee records into a public chatbot. These materials may include personal details or sensitive context that employees did not expect to be shared outside company systems.
If you need help drafting a job description or a general HR message, use a fictional example or a blank template rather than a real employee’s information.
Passwords, access codes, and security details
Never enter passwords, authentication codes, private keys, recovery answers, API keys, or other credentials. Keep out security configurations and details that could expose how your organization’s systems are accessed or protected. A chatbot is not a password manager or a secure place to troubleshoot account access.
Financial and payment information
Keep bank details, payment card information, payroll records, tax documents, invoices tied to identifiable customers, budgets, forecasts, and nonpublic financial reports out of public tools. Even when a task seems routine, a spreadsheet or copied email may contain more than the figures you intended to share.
Legal, confidential, or privileged material
Do not submit contracts, legal advice, dispute details, investigation notes, confidential correspondence, or material marked privileged or confidential unless your organization has specifically approved the tool and the use. For legal and regulated work, follow the relevant professional and organizational requirements rather than relying on a general chatbot rule.
Trade secrets and internal business information
Keep product plans, pricing strategies, designs, source code, unpublished research, customer lists, internal reports, and other proprietary material out of public chatbots. The same caution applies to operational details that are not public, such as internal processes, network diagrams, incident reports, and business continuity plans.
Small pieces of context can add up. A prompt may not include a company name but could reveal it through a project description, a named location, a distinctive product feature, or details copied from an internal discussion.
Check more than the prompt text
Information can be exposed indirectly. Before using a public chatbot, look for sensitive details in:
- Email threads, including forwarded messages and signatures.
- Documents, spreadsheets, and presentation notes.
- Screenshots or copied text that show names, account details, or internal systems.
- Source code, configuration files, logs, and error messages.
- Document properties or other metadata that may identify people or projects.
- Examples that combine details and could identify someone, even without a name.
Do not assume that a file is safe because the portion you want help with looks harmless. Review the entire material before sharing it, and avoid uploading a complete document when a generic description would do.
Safer ways to get help with common tasks
Many everyday tasks can be described without using confidential information. For instance, ask for a general outline, a neutral rewrite, or a sample process using fictional details. Replace real people, customers, vendors, products, and figures with clearly invented placeholders.
Share only the minimum context needed to get useful help. Instead of pasting a real customer complaint, describe the issue in broad terms and ask for a professional response structure. Instead of providing internal code or logs, ask a general troubleshooting question without including secrets, system identifiers, or private configuration details.
De-identification can reduce exposure, but it is not a guarantee. Details may still identify someone when combined, and removing names does not necessarily remove confidential business information. If a task genuinely requires sensitive data, use only an organization-approved environment and confirm that the specific information and purpose are permitted by policy.
Public tools and approved workplace tools are not the same
Public consumer chatbots and organization-approved AI tools may have different terms, settings, access controls, and data-handling practices. Do not assume that a tool will keep prompts private, delete them, or prevent their use in other ways. Check the organization’s approved-tool list and guidance, and ask the designated security, privacy, or IT contact if anything is unclear.
An employee AI data-handling policy should explain which tools are allowed, what data categories can be used, what review is needed, and where to report concerns. It should also make clear that approval for one task or type of information does not authorize all uses. Businesses reviewing their broader technology practices can also explore IT support considerations for Florida businesses.
What to do if confidential information was shared
If an employee accidentally submits sensitive information, respond promptly and follow internal procedures. Avoid making the situation worse by submitting more data or trying additional actions without guidance.
- Stop further sharing. Do not continue the conversation with more confidential details.
- Preserve relevant details. Note what was shared, when it happened, which tool was used, and any available information about the prompt or file. Do not circulate the sensitive material more widely.
- Notify the designated contact. Tell your manager or the organization’s security, privacy, or IT contact as directed by policy.
- Follow internal instructions. The organization can assess the situation and decide what steps are appropriate under its procedures and obligations.
Employees should not assume that deleting a prompt from a visible chat removes all copies or resolves the incident. Follow the organization’s direction and let the appropriate team assess what happened.
Frequently asked questions
Can employees paste customer or patient information into a public AI chatbot?
As a general workplace rule, do not. Use sensitive records only in an environment and for a purpose your organization has specifically approved, and follow applicable requirements.
Is it safe to upload company documents to a public AI tool?
Not by default. Documents may contain confidential content, personal information, credentials, or metadata. Check the tool and document against company policy before uploading anything.
Can public AI chatbots store or reuse prompts?
Practices and settings can differ by tool. Do not assume prompts will remain private or will not be retained or used in other ways. Check the tool’s terms and your organization’s guidance.
How can a business let employees use AI while protecting sensitive data?
Set clear rules for approved tools and data types, provide safe examples, and establish a simple process for questions and accidental disclosures. A broader review of technology practices may also benefit from guidance on business productivity and IT support.
Make safe AI use part of everyday work
The practical rule is to keep customer, employee, financial, legal, credential, and nonpublic operational information out of public AI chatbots unless the organization has explicitly approved the tool and use. Clear examples and an easy escalation path help employees make that decision before sensitive information is shared.
Zevonix provides managed IT support and technology services to businesses across Florida, including cybersecurity and compliance support. If your organization is reviewing how employees use AI tools, learn about business IT support options or contact Zevonix to discuss your needs.