The Privacy Rule: What Stays Off Public AI
Module 2 — Responsible AI at Work | Priority for knowledge workers, analysts, HR, finance, and anyone handling confidential data
The Privacy Rule is simple: If it’s confidential, it stays off public AI. No exceptions.
The challenge is that confidential information rarely announces itself in the moment. A ten-page meeting transcript doesn’t feel like a trade secret — it feels like notes that need summarizing. A client’s name in a prompt doesn’t feel like a GDPR violation — it feels like context. A financial model pasted into an AI tool doesn’t feel like a data breach — it feels like a time-saving shortcut.
This module teaches employees to recognize the difference between how confidential data feels in the moment and the risk it poses the instant it enters a public AI tool.
What This Module Covers
Risks This Module Addresses
- Client names, financial data, and strategy documents entering public AI tools via routine prompts
- GDPR violations from personal data submitted to public AI processors without consent
- Trade secret exposure through confidential documents pasted into ChatGPT, Gemini, or Copilot
- Employees unaware of the difference between public AI tools and company-approved sandbox environments
- No practical sanitization habit — employees either avoid AI entirely or use it without boundaries
Course Scenarios — Privacy and Data Exposure
Module 2 addresses the moment where confidential data feels safe to share because prompting feels private. The following scenario is representative of the decision situations employees work through in the course.
📽 From the Course — Scenario: Data Exposure and AI Prompts
The Competitive Intelligence Risk
Imagine if a competitor asked an AI tool: “What is [Our Company’s] strategy for next year?” — and the AI answered using the exact notes your employee just uploaded. Public AI models can and do incorporate user input into the knowledge they draw on. The information you enter does not persist across sessions.
The employee who pasted the meeting transcript had no intent to disclose. Intent is irrelevant. The data left the organization the moment it entered the prompt field.
Course Reference — What Belongs Off Public AI
The course provides employees with a clear reference framework covering the most common data types that pose exposure risk, why each is dangerous in a public AI prompt, and the safe alternative for each situation.
| Data Type | Risk if Entered in Public AI | Safe Alternative |
|---|---|---|
| Client names and contact data | GDPR violation — personal data of a third party disclosed to a public processor without consent | Remove all names before prompting — describe as “a client in [industry]” |
| Internal financial data | Confidentiality breach — unreleased financial information may be retained by the model | Use only publicly available figures. Use the approved sandbox for internal analysis. |
| Proprietary code or algorithms | IP and security risk — code patterns may be retained, vulnerabilities may be introduced | Use company-approved AI coding tools with data isolation enabled |
| Strategy documents and roadmaps | Trade secret disclosure — competitive intelligence may be accessible to others via the model | Strip all identifying details. Summarize manually before prompting. |
| Employee performance data | Privacy violation — personal employee information is regulated in most jurisdictions | Never enter employee data into public AI tools. Use HR-approved tools only. |
| Legal and regulatory documents | Privilege waiver risk and confidentiality breach | Legal documents require legal team review before any AI processing. |
Course Technique — The Sanitization Method
The course teaches employees a practical technique for using an AI productively when the raw materials contain confidential information. The sanitization method — removing or replacing all identifying details before prompting — takes sixty seconds and eliminates the exposure. The following examples show exactly how employees apply it.
| ✗ Original (Do Not Enter) | ✓ Sanitized (Safe to Enter) |
|---|---|
| Summarize this Q3 strategy deck from our Acme Corp client meeting | Summarize this Q3 strategy document for a financial services client |
| Help me draft a response to John Martinez at NovaTech about the contract delay | Help me draft a professional response to a client about a contract delay |
| Here is our unreleased product roadmap for 2026 — generate talking points | Generate talking points for a new software product launching in mid-year |
| Review this Python code from our proprietary trading algorithm | Review this Python code for logical errors — [replace proprietary logic with generic equivalent] |
Course Assessment — Knowledge Checks Throughout
Each module tests employee understanding through scenario-based knowledge checks that require applying the privacy rule — not just recalling definitions. The following example reflects the format and difficulty level employees encounter in Module 2.
📝 From the Course — Knowledge Check
The Quarterly Strategy Meeting Transcript
You have a transcript from your company’s quarterly strategy meeting — ten pages covering the unreleased product roadmap, Q3 financial targets, and a pending acquisition discussion. Your manager needs a bulleted summary sent to the team today.
What is the right course of action?
❌ Option A — Paste the full transcript into ChatGPT for a quick summary
❌ Option B — Use the sanitization technique to remove sensitive details, then prompt
✅ Option C — Use the company-approved sandbox AI tool, or summarize manually
Why C is correct: The transcript contains trade secrets and financial data that are integral to the document — not incidental. Sanitization is not appropriate here because removing the sensitive content would remove the substance. Use a sandboxed tool or summarize manually.
Reinforcing the Privacy Rule Throughout the Year
Annual training establishes the rule. The pause reflex — the moment an employee asks “is this data private?” before reaching the prompt field — fades as AI use grows and new tools appear. Xcelus addresses this through the Compliance Reinforcement Cycle™ — short scenario reminders deployed throughout the year that keep the privacy rule top of mind when employees need it most.
Course Deployment Information
| 📋 Module 2 — At a Glance | |
|---|---|
| Audience | Knowledge workers, marketers, analysts, HR, and finance teams |
| Length | 15–20 minutes |
| Format | Scenario-based, SCORM-compatible, LMS-ready |
| Covers | Confidential data in AI prompts, the sanitization technique, GDPR intersection, and company-approved sandbox tools |
| Prerequisite | Module 1 — AI Fundamentals recommended |
| Regulatory alignment | GDPR, EU AI Act, UK AI Principles |
Frequently Asked Questions
Platform terms change, and the absence of training-use claims does not guarantee data isolation. Company policy applies regardless of what an individual platform’s terms say. For confidential data, use company-approved sandbox tools — not public tools, regardless of their stated data practices.
If information is genuinely public — a company press release, a published annual report, publicly available regulatory filings — it can generally be entered into a public AI tool. The rule applies to confidential, internal, or personal data. When in doubt, treat it as confidential.
Integrated AI tools in enterprise software often have different data handling arrangements than standalone public tools. Your IT or legal team will specify which integrated tools are approved and under what conditions. Do not assume that an AI tool built into approved software automatically handles your data safely — verify with your IT team.
GDPR violations — including inadvertent disclosure of personal data to unauthorized processors — can result in regulatory investigation, significant fines, and notification obligations. The fact that the disclosure was unintentional does not eliminate the obligation or the consequence.
Most employees encounter more confidential information than they realize — client names in emails, internal project details, colleague information. The training is designed to help all employees recognize when the information they are working with crosses into territory that requires more care before prompting.
Continue the Series
|
Responsible AI at Work — 4-Module Series ▶ Module 2: The Privacy Rule — What Stays Off Public AI (this page) Module 3: The Accountability Rule |
Ready to Deploy Module 2?
Module 2 can be deployed as a standalone course for high-risk employee groups or as part of the full Responsible AI at Work series. SCORM-compatible and LMS-ready. Scenarios can be customized to reflect your organization’s approved AI tools, data classification policies, and sandbox environment.