When deciding what company information employees should never paste into public AI tools, use a simple starting point: keep out anything restricted by law, contract, client duty, or company policy, or anything that could cause harm if exposed. Only use that information with a specific AI service and use case your organization has approved.
This rule covers more than obvious secrets. A customer email thread, payroll export, contract draft, source-code snippet, or screenshot may contain details that should not leave approved systems. And removing a name does not always make information anonymous if other details could still identify the person or business.
Information to keep out of public or unapproved AI
- Customer, client, and employee records: Names paired with email addresses, phone numbers, account details, customer correspondence, personnel files, performance notes, and hiring records. Avoid uploading a full email thread or spreadsheet just to ask for a summary.
- Health, student, and other regulated records: Patient information, student records, and sensitive records handled under industry-specific rules, contracts, or professional duties. The U.S. Department of Education’s guidance on student privacy and online educational services is a useful reminder that organizations should consider privacy when using digital tools with student information.
- Credentials and access details: Passwords, API keys, authentication tokens, recovery codes, private keys, and verification codes. Do not paste them into a chatbot to troubleshoot an account or generate a script.
- Confidential communications and documents: Client messages, legal correspondence, contract drafts, negotiation notes, and documents shared under confidentiality obligations. A request to “summarize this contract” can still expose its contents.
- Financial and payroll information: Bank or payment details, payroll exports, tax records, financial statements, budgets, and nonpublic forecasts. Remove more than names if other fields could reveal an individual’s identity or a company’s plans.
- Business secrets and unreleased work: Source code, product plans, designs, internal procedures, security configurations, pricing plans, and research that the business has not made public. Treat screenshots and copied snippets as carefully as complete files.
Public AI and approved AI are not automatically the same
A tool’s label does not tell you whether it is suitable for confidential work. Public tools and personal accounts may handle prompts differently from organization-approved services. Before approving a tool, the business should review its data handling and retention, access controls, contractual terms, and fit for the intended use. It should also consider any legal, client, or professional obligations that apply to the information.
Make approval specific. A tool allowed for drafting a public-facing announcement is not automatically approved for summarizing a customer file. If the intended task or information changes, check the rules again. For legal work, for example, the Florida Bar’s Opinion 24-1 addresses professional considerations related to lawyers’ use of generative AI. It is not a substitute for an organization’s own review of its obligations.
Give employees a decision rule they can use
Policies work better when people can apply them in the moment. Ask employees to pause before entering work information into an AI tool and consider:
- Is this information confidential, personal, regulated, contract-restricted, or limited by a client or professional duty?
- Could someone identify a customer, patient, employee, student, or client from the details, even if a name is removed?
- Has the organization approved this specific tool for this kind of information and this task?
- Can I complete the task with fictional details, less context, or an approved internal process instead?
If the employee is unsure, the answer should be to stop and ask the designated manager, privacy contact, or security contact. Do not rely on a coworker’s past use of a tool as proof that a new task is approved.
Offer a safer way to get the work done
A useful AI rule should explain what employees can do, not only what they cannot do. For brainstorming or formatting, replace real customer details with a fictional example. For summarization, use an approved service and provide only the minimum information needed. For a writing task, describe the situation in general terms rather than copying a real client message.
De-identification takes care. Removing a name may not be enough when a job title, location, unusual event, account detail, or combination of facts could point to a particular person or organization. If details still feel recognizable to someone familiar with the situation, treat them as sensitive.
Owners should make the rules easy to find and explain how employees can request approval. Include examples drawn from everyday work, such as customer email threads, payroll files, draft agreements, code, and screenshots. State who approves tools and use cases, which data types are off limits, and who to contact with questions.
If someone pastes confidential information by mistake
Tell employees to act promptly rather than assume deleting the chat resolves the exposure. They should stop sharing or using the material, notify the designated internal contact, preserve relevant details about what was entered and where, and follow the organization’s incident-response process. The appropriate next steps depend on the information, the tool, and the organization’s obligations, so employees should not make their own assumptions about whether the incident is harmless.
Clear ownership matters here. If employees are unsure who handles security questions or how to report an issue, reviewing common business IT support concerns can help frame what to clarify with an IT provider. If responsibility for IT support is already under review, this guide to changing IT providers may also help you think through continuity and accountability.
Frequently asked questions
Can employees paste customer names, emails, or account records into a chatbot?
Not unless the organization has approved that specific tool and use for that information. A name, email, account number, or message may identify a customer or reveal confidential details.
Is it safe to enter patient information or other health records into public AI?
Do not enter it into a public or unapproved tool. Health information may carry legal, contractual, or professional handling requirements. Ask the organization’s privacy or security contact which tools and workflows, if any, are approved.
Can employees use public AI to summarize confidential contracts or legal documents?
Not unless the organization has approved the tool for that document and task. Summarizing still requires sharing the document’s contents. Check applicable confidentiality duties and internal rules first.
What should schools and employers consider before using AI with student or employee records?
Review the tool’s data handling, retention, access controls, and terms, along with the organization’s obligations. Decide which records may be used, for what purpose, and who can approve the workflow.
What should an employee do after accidentally pasting confidential information into AI?
Stop sharing it, promptly notify the designated internal contact, preserve relevant details, and follow the organization’s response process. Do not assume deleting the chat settles the issue.
Set the rule before the next prompt
Start with a short, plain-language list of information employees must keep out of public or unapproved AI, then name the person or team that can approve tools and answer questions. Zevonix offers managed AI services alongside its managed IT and cybersecurity services for businesses in Palm Coast and Flagler County. To discuss your organization’s IT needs, contact Zevonix at (386) 260-0777.