Think Before You Prompt: Workplace Chatbots and Confidential Information

Written on behalf of Peter McSherry

Artificial intelligence chatbots are becoming part of everyday workplace activity. Employees may use them to summarize documents, draft correspondence, prepare presentations, analyze data, create meeting notes, or generate ideas. These tools can save time, but their convenience may encourage users to enter information without considering where it goes or how it may be handled.

A chatbot prompt can contain far more than a simple instruction. It may include customer details, employee records, internal financial information, legal correspondence, proprietary methods, or material relating to a pending transaction. Once information is entered into an external system, the employer may have limited control over its storage, processing, retention, or future use.

For Guelph employers, the increasing use of workplace chatbots raises questions about confidentiality, privacy, cybersecurity, employee responsibilities, and internal governance. These issues can arise even when employees use AI tools with productive intent.

What Information Might Employees Enter Into a Chatbot?

Employees may disclose confidential information unintentionally when they paste source material into a chatbot. For example, an employee asking for help drafting a response to a customer may include the customer’s name, contact information, account history, complaint, or payment details.

Other chatbot prompts may contain employee performance records, accommodation information, disciplinary documents, compensation data, business plans, product specifications, source code, pricing strategies, supplier terms, or unpublished financial results. Even a short excerpt may reveal sensitive information when combined with other details.

The Canadian Centre for Cyber Security has warned that users may unknowingly provide sensitive corporate data or personally identifiable information through AI prompts. It has also identified the risk that employees may disclose personal or confidential commercial information outside approved security and policy frameworks.

Why Chatbot Inputs Can Create Confidentiality Risks

Employees sometimes treat a chatbot like an internal search engine or private digital assistant. However, a public chatbot is generally operated by an external service provider. The employer may not control the provider’s systems, subcontractors, server locations, retention practices, or security arrangements.

The applicable terms may also differ between free consumer accounts and paid enterprise products. Depending on the product and settings, prompts may be retained, reviewed for safety or quality purposes, or used to improve services. Employers may therefore need to understand the specific tool being used rather than relying on general assumptions about AI platforms.

Removing names may not always make information anonymous. A description of a person’s role, workplace, medical circumstances, transaction, or dispute could still allow that individual or matter to be identified. Confidential information may also remain commercially sensitive even when it does not identify a person.

Confidentiality Responsibilities Still Apply

The use of a new technology does not necessarily replace existing workplace responsibilities. Employees may already be subject to confidentiality obligations arising from employment agreements, workplace policies, professional requirements, or the nature of their positions.

An employee who enters restricted information into an unauthorized chatbot may create concerns similar to those arising from sending documents to a personal email address or storing records in an unapproved cloud service. The relevant circumstances may include the type of information involved, the employee’s role, the employer’s policies, the tool’s terms, and whether the disclosure was intentional.

Employers may also have confidentiality commitments to customers, vendors, business partners, investors, and other third parties. A chatbot disclosure could therefore affect both contractual relationships and the employment relationship.

Personal Information Requires Additional Attention

Chatbot prompts can involve personal information when employees use AI to review resumés, summarize customer files, prepare performance documentation, analyze complaints, or draft communications about identifiable individuals.

Canadian privacy regulators have stated that existing privacy laws continue to apply to the development and use of generative AI. Their guidance emphasizes accountability, appropriate purposes, consent where required, limiting collection and use, safeguards, transparency, and careful consideration of whether personal information should be entered into an AI system.

The precise legal framework depends on factors such as the employer’s industry, the information involved, and how it is collected, used, or disclosed. PIPEDA applies to private-sector organizations that handle personal information in the course of commercial activities, although its treatment of employee information depends partly on whether the organization operates a federal work, undertaking, or business.

The Difference Between Approved and “Shadow” AI

Some organizations provide employees with approved enterprise AI systems that include negotiated contractual protections, administrative controls, access restrictions, logging, and settings designed to limit data retention or model training.

Other employees may independently create accounts with publicly available chatbots because those tools are familiar or easy to access. This unauthorized use is sometimes described as “shadow AI.” It may occur without the knowledge of the employer’s information technology, privacy, legal, or human resources teams.

A complete prohibition may not eliminate shadow AI if employees believe the technology is necessary for efficient work. Employers may instead consider whether clear rules, approved alternatives, practical examples, and accessible guidance could reduce the likelihood of undisclosed use.

What a Workplace AI Policy May Address

A workplace AI policy can identify which tools are approved, which tasks are permitted, and which categories of information must not be entered into a chatbot. The policy may distinguish between public tools, enterprise systems, internally hosted models, and AI features built into existing workplace software.

It may also explain whether employees can use chatbots to prepare customer communications, employment documents, contracts, reports, code, research, or decision-making materials. Where certain uses require approval, the policy can identify who provides it and how employees should document their use of the technology.

Examples are especially important. Telling employees not to disclose “confidential information” may be too general if they do not understand that the term includes customer lists, resumés, internal emails, meeting transcripts, unreleased financial data, business strategies, passwords, and details about workplace investigations.

Human Review Remains Important

Confidentiality is not the only concern associated with chatbot use. AI-generated material may be incomplete, inaccurate, outdated, or unsuitable for the intended audience. An employee may also receive an answer that appears authoritative even though the output does not reliably reflect the source material.

Employers may establish review requirements for AI-assisted work. Depending on the context, employees may be expected to verify facts, review citations, remove inappropriate language, check calculations, confirm that the output reflects company policy, and ensure that a qualified person remains responsible for the final decision.

The Office of the Information and Privacy Commissioner of Ontario and the Ontario Human Rights Commission have identified privacy protection, transparency, accountability, safety, reliability, and human rights as principles for responsible AI use.

Training Can Make Policies More Effective

Employees may not realize that pasting text into a chatbot constitutes disclosure. Training can demonstrate how a seemingly ordinary prompt may expose personal, privileged, proprietary, or commercially sensitive information.

Practical exercises could show employees how to use fictional examples, remove unnecessary identifiers, summarize information before entering it, or use an approved internal tool instead. Training may also explain how to recognize confidential information and whom to contact when an employee is unsure.

Because AI products and workplace uses continue to change, employers may periodically update their training and policies. The Canadian Centre for Cyber Security recommends employee training on protecting sensitive information and reporting security incidents.

Responding to an Accidental Disclosure

An employee who realizes that confidential information has been entered into an unauthorized chatbot may be reluctant to report it. A reporting process can encourage employees to raise issues promptly so the organization can assess which information was involved and what steps may be available.

The response may include preserving relevant records, reviewing the provider’s settings and contractual terms, requesting deletion where possible, changing compromised credentials, notifying internal privacy or security personnel, and assessing whether any contractual or statutory notification obligations are engaged.

The organization may also review why the disclosure occurred. An incident could reveal unclear policies, inadequate training, a lack of approved tools, excessive access permissions, or a workflow that encourages employees to use external services.

Building a Practical Approach to Workplace Chatbots

Workplace chatbots can offer meaningful administrative and productivity benefits, but they do not eliminate the need to protect confidential information. Employers may need to evaluate tools before deployment, map how information moves through them, and decide what categories of data are appropriate for AI-assisted work.

Employees also benefit from knowing the boundaries before using a chatbot. Clear policies can explain not only what is prohibited, but also how approved AI tools may be used safely and who can answer questions.

As AI becomes more deeply integrated into common business software, chatbot governance may be incorporated into broader confidentiality, privacy, cybersecurity, records management, and workplace policy frameworks. A coordinated approach can help organizations address the technology without separating it from their existing responsibilities.

Peter A. McSherry Employment Lawyers: Guidance for Workplace AI Policies

The use of workplace chatbots can affect confidentiality policies, employment agreements, privacy practices, discipline, investigations, cybersecurity procedures, and employee training. At Peter A. McSherry Employment Lawyers, our employment lawyers assist employers across Guelph and the surrounding area with reviewing workplace AI policies, updating confidentiality provisions, assessing employee use of generative AI, and responding to potential information disclosures.

Contact our Guelph employment law firm at 519-821-5465 or visit us online to discuss workplace chatbots, confidential business information, employee privacy, AI governance, or related employment policy considerations in Ontario.