Public AI chatbots can help employees brainstorm, rewrite general text, or organize ideas. But a prompt can also expose sensitive business information, especially when someone pastes a full email thread, uploads a document, or includes details that identify a customer or employee.
The practical rule is simple: if information is private, confidential, or entrusted to your business, do not put it into a public chatbot unless your organization has approved that specific tool and its use for that specific information. The U.S. Small Business Administration’s guidance on AI for small businesses encourages businesses to consider AI risks as part of responsible adoption.
Here is a clear stop-list for employees, along with safer ways to use AI for routine tasks.
Information employees should keep out of public AI chatbots
Do not paste or upload the following into a public tool that your employer has not approved for that data:
- Passwords, access tokens, and security details. Keep passwords, recovery codes, API keys, private keys, authentication links, and system configuration details out of prompts. These can provide a path to business accounts or systems.
- Customer or client records. Avoid names paired with email addresses, phone numbers, account numbers, order details, support histories, or other information that could identify a person or reveal their relationship with your business.
- Employee and applicant information. Do not submit personnel files, payroll details, performance notes, disciplinary records, job applications, interview notes, or other private workplace information.
- Health information. Keep patient details, medical records, appointment information, and other health-related data out of unapproved public tools. Healthcare organizations should follow their own privacy and security procedures and confirm whether a tool is authorized for the intended use.
- Financial and payment information. Do not paste bank details, payment card data, tax records, financial statements, invoices containing personal information, or account credentials into an unapproved chatbot.
- Government identifiers. Keep Social Security numbers, driver’s license details, passport information, and other government-issued identifiers out of prompts and uploads.
- Confidential company material. This can include contracts, legal correspondence, pricing plans, budgets, sales forecasts, source code, product designs, internal incident reports, and business plans that have not been made public.
These categories are a starting point, not an exhaustive inventory. A seemingly ordinary detail can become sensitive when combined with other information. For example, a project description that omits a customer’s name may still identify that customer through a unique location, timeline, or set of circumstances.
Remember that files and context can disclose more than a prompt
Employees may recognize that a spreadsheet contains private information but overlook the same risk in an email thread, meeting notes, screenshot, or copied support ticket. Attachments can include names, contact details, account identifiers, internal comments, or document properties that are easy to miss.
Before sharing any material with an AI tool, consider the whole submission: the prompt, the file, the surrounding context, and any details that could be combined to identify a person, customer, project, or internal process. Removing a name alone may not be enough to make the information anonymous.
For a useful decision point, ask: would the company be comfortable if this information were seen outside the organization? If the answer is no, do not enter it into a public chatbot unless the tool and use have been specifically approved.
Public chatbot or employer-approved tool?
Not every AI tool has the same privacy and security terms. A tool selected and configured by an employer may offer different controls from a public consumer chatbot, but employees should not assume that a business account automatically makes every use appropriate.
Before approving a tool for work, the responsible people in your organization should review its actual terms and settings. Check how prompts and uploaded files are handled, whether they may be retained or reviewed, who can access them, what deletion controls are available, and whether the tool is approved for the kinds of information employees plan to use. The NIST Generative AI Risk Management Framework profile offers a structured resource for considering risks associated with generative AI.
A setting that limits model training does not, by itself, answer every data-handling question. Retention, access, review, deletion, and contractual terms still matter. Do not treat a general product assurance or a single privacy setting as permission to submit confidential material.
Businesses in healthcare, finance, legal, education, government, and other sensitive fields may also have contractual, organizational, or regulatory requirements to consider. When in doubt, employees should pause and ask the person responsible for privacy, security, or compliance rather than testing a tool with real records.
Safer ways to use AI for routine work
Many everyday tasks can be completed without sharing real private data. Employees can make a prompt safer by reducing the information it contains:
- Use a fictional example. Ask for a sample response using invented names and details rather than copying a real customer or employee case.
- Remove identifiers and unnecessary context. Replace names and account details with generic labels, and omit background information that is not needed to complete the task.
- Share only the minimum needed. If the goal is to improve the tone of a message, provide a short, sanitized excerpt rather than an entire email chain or file.
- Keep sensitive work in approved systems. If the task requires real customer, employee, financial, or confidential company information, use only a tool and workflow authorized by your organization.
- Review the result before using it. Check AI-generated work for accuracy and make sure it does not reveal confidential details or introduce unsupported claims.
For instance, instead of pasting a customer complaint with contact information, an employee could describe the issue in general terms and ask for a neutral outline of a response. Instead of uploading a real internal report, they could ask for a template using fictional content.
These practices work best when employees have clear written guidance. A useful AI acceptable-use policy can explain which tools are approved, what information is off-limits, who can authorize exceptions, and how staff should report a mistake. Pair the policy with practical employee cybersecurity education so workers can recognize accidental disclosures in prompts, documents, and attachments.
What to do if sensitive information was submitted
If an employee realizes they entered private information into an unapproved chatbot, they should promptly tell the appropriate manager or person responsible for IT, security, or privacy. Follow the organization’s incident-reporting process rather than trying to handle the situation quietly.
Record what was submitted, which tool was used, and when it happened, without copying the sensitive information into additional messages. The responsible team can review the tool’s available controls and terms, determine what follow-up is appropriate, and decide whether other steps are needed. Deleting a prompt from a visible chat history should not be treated as proof that all copies or records have been removed.
Frequently asked questions
Is it safe to paste customer names, emails, or account details into an AI chatbot?
Do not enter identifiable customer information into a public chatbot unless your organization has approved that tool and use for that data. For ordinary drafting or brainstorming, use fictional details or a sanitized summary instead.
Can employees use public AI chatbots to summarize confidential company documents?
Not unless the organization has specifically approved the tool and workflow for those documents. A file may contain sensitive information in its text, attachments, comments, or context, even if the request is only to create a summary.
Should staff enter employee records or job applicant information into AI tools?
Keep those records out of public chatbots. Use only an employer-approved system for an authorized purpose, and follow your organization’s procedures for handling personnel and applicant information.
Does turning off AI training make sensitive data safe to submit?
No single setting answers every privacy question. Check the tool’s retention, access, review, deletion, and contractual terms, and follow your employer’s approval process before using sensitive data.
Make safe AI use an everyday habit
The safest public-chatbot prompt is one that does not contain information your business needs to protect. Give employees a clear stop-list, approved options, and a straightforward way to ask for guidance when a task involves real customer, employee, financial, or confidential company information.
For businesses in Palm Coast and Flagler County, Zevonix provides proactive managed IT services and cybersecurity support. To discuss practical technology safeguards for your business, contact Zevonix at (386) 260-0777.