Can employees put confidential information into public AI chatbots? In general, they should not enter client, patient, financial, or other sensitive business information unless the organization has approved that specific tool and confirmed that its settings and terms are appropriate for the information involved.
The word “public” does not necessarily mean everyone can see a prompt. It often describes a consumer-facing service rather than a tool managed by your organization. But the service may handle prompts under its own rules for retention, access, sharing, and model improvement. Those details matter, and they can differ between products and account types.
A simple policy can help employees move quickly without guessing: use public or approved information freely, use internal information only in an approved environment, and keep highly sensitive data out unless a responsible team has explicitly approved the use.
Use a simple stoplight rule
Green: Public or already-approved material
Examples include information published on your website, public product descriptions, or a generic request to improve the tone of a non-confidential message. Employees should still follow company policy and review chatbot output before using it, but these prompts generally do not expose private business details.
Yellow: Internal information that needs an approved tool
Internal procedures, nonpublic project notes, draft communications, and operational details may be useful in AI-assisted work. They should go into a tool only if your organization has approved the specific service and use case. Check what account is being used, which privacy settings apply, who can access the content, and whether the organization has reviewed the service terms.
When possible, reduce the prompt to the minimum needed. Replace real names and details with placeholders, or use a fictional example that preserves the task without exposing the underlying information.
Red: Do not paste into an unapproved public chatbot
Keep these categories out of consumer-facing tools unless your organization has specifically approved the exact tool and safeguards for that data:
- Passwords, authentication codes, API keys, recovery answers, or other access credentials.
- Identifiable patient or client records, including notes, case details, contact information, and records that combine several identifying details.
- Payment card information, bank details, account numbers, tax information, or other financial records tied to a person or business.
- Confidential contracts, legal matter details, personnel records, or private customer communications.
- Proprietary designs, source code, unpublished plans, pricing strategy, or other information that could harm the organization if disclosed.
Removing a name does not automatically make information anonymous. A location, role, unusual event, date, or combination of details can still point to a particular person or organization. If an employee cannot confidently tell whether a prompt could identify someone or reveal protected information, they should leave it out and ask the appropriate internal contact.
Why “don’t train on my prompts” is not the whole answer
Some services provide settings that limit whether submitted content is used to improve models. That setting is worth reviewing, but it does not answer every privacy question. It may not describe how long content is retained, whether authorized personnel can access it, what sharing features are enabled, or how the service may use data under its terms.
Review the actual product, account type, settings, and agreement. Do not assume that a setting in one service applies to another, or that deleting a conversation resolves every concern. The National Institute of Standards and Technology identifies data protection and retention as considerations in managing generative AI risks; see its Generative AI Risk Management Profile.
What healthcare and finance teams should check
Healthcare organizations should not rely on a chatbot’s general privacy claims to decide whether it is appropriate for protected health information. Before using a tool with that information, confirm that the specific product and configuration have been reviewed, that required agreements are in place, and that the use fits the organization’s HIPAA risk analysis and policies. If those checks have not been completed, keep identifiable patient information out.
Financial and other businesses should take a similar approach with account details, customer records, payment information, and confidential business documents. Check the organization’s data-handling rules and any obligations that apply to the information. When the status is unclear, ask the privacy, compliance, or security lead before using the chatbot.
Safer ways to get value from AI at work
Businesses do not have to choose between banning every AI use and allowing employees to paste anything into a prompt. A controlled approach gives staff practical options while reducing unnecessary exposure:
- Choose an approved service. Identify which tools and account types are allowed, and what kinds of work each one may support.
- Minimize the input. Include only the information needed to complete the task. Remove names, account details, and other identifiers when they are not essential.
- Use synthetic examples. For drafting, testing, or summarizing workflows, use made-up details rather than actual client or patient records.
- Check retention and access. Review the applicable settings and terms, including how submitted content may be stored, accessed, shared, or used.
- Set clear boundaries. An AI acceptable use policy should explain approved tools, permitted tasks, prohibited data, and where employees can ask questions.
- Keep people accountable. Employees should verify AI output before relying on it, especially when it could affect customers, patients, finances, or business decisions.
For broader security practices that help protect business information, see Zevonix’s guide to cybersecurity for Jacksonville businesses. Clear access controls, employee guidance, and a process for reporting concerns can support responsible technology use.
If someone already entered sensitive information
Respond promptly, without making the situation harder to assess. The employee should stop using that chat for sensitive work, avoid pasting additional details, and preserve relevant information such as the service and account used, the approximate time, and the type of data submitted. They should notify the organization’s privacy or security contact as soon as possible.
The responsible team can review the service terms and settings, assess what information may have been exposed, and determine what steps are appropriate. Employees should not assume that deleting a conversation settles the issue, and should follow the organization’s incident-reporting process rather than trying to handle it privately.
Frequently asked questions
Can employees enter client or patient information into a public AI chatbot?
They should not do so unless the organization has specifically approved the tool, configuration, and use for that information. A consumer-facing chatbot should not be treated as an approved place for confidential records by default.
Is turning off model training enough to make a chatbot safe for confidential information?
No. Also review retention, access, sharing, account settings, service terms, and company policy. A training setting alone does not establish that a tool is appropriate for sensitive data.
Is removing a name enough to anonymize a record?
Not necessarily. Other details, or a combination of details, may still identify a person or client. Use synthetic examples or ask the appropriate internal team to review the data.
What should a healthcare organization check before using AI with protected health information?
Review the exact service and configuration, applicable agreements, access and retention practices, and the organization’s HIPAA risk analysis and policies before using protected information.
What should an employee do after entering sensitive information into a chatbot?
Stop using the chat for sensitive work, preserve relevant details, and promptly notify the organization’s privacy or security contact. Follow the organization’s incident process.
Make the safe choice easy
Employees need a clear answer before they are staring at a prompt with a client record or business document ready to paste. Define which tools are approved, state what information must stay out, and give staff a straightforward way to ask for help.
Zevonix provides managed IT and cybersecurity services to Jacksonville businesses. If your organization needs support reviewing its technology practices, you can learn about managed IT services for Jacksonville businesses or contact Zevonix to discuss your needs.