Organisations should teach employees specific rules for what can and cannot be entered into public AI tools, then reinforce those rules with role-based scenarios. Training should cover confidential data, customer information, credentials, and intellectual property. The goal is not to ban AI use, but to reduce leakage risk through clear guidance, just-in-time nudges, and consistent governance.
What employees need to learn before using public AI tools
Training should start with a simple rule set, not a blanket warning. Employees need to know which data classes are prohibited, which are allowed only in redacted form, and which requests must be handled through approved internal tools instead. That distinction matters because public AI tools can retain prompts, surface them in logs, or use them in ways the user did not intend.
Teach people to recognise sensitive inputs in everyday work: customer records, internal strategy, source code, credentials, incident details, legal material, and commercially sensitive plans. The practical test is whether the text would still be safe if copied into an external system, reviewed by a third party, or stored longer than expected. If the answer is no, it should not go into a public AI tool.
Employees also need examples of safe substitutes. A good training module shows how to ask for structure, summarisation, drafting, translation, or brainstorming without including raw sensitive content. The point is to preserve the productivity value of AI while changing the input pattern, so staff learn how to reframe a task rather than abandon it.
How role-based scenarios make the rules usable
Role-based scenarios turn policy into judgement. A software engineer, finance analyst, HR partner, sales rep, and support agent each face different data boundaries, so the same rule set should be demonstrated through situations they actually recognise. This is where the training becomes durable, because employees are more likely to remember a decision they practised than a policy line they skimmed.
Scenario design should include common edge cases, not only obvious violations. For example, a user may want to paste an error trace that includes tokens, a spreadsheet that contains customer identifiers, or a meeting note that mixes public and confidential details. Training should show how to strip the sensitive layer, use a company-approved environment, or ask for help before submitting the prompt.
A useful pattern is to tie each scenario to a decision rule: if the prompt contains regulated, confidential, or business-critical information, do not use a public tool; if the task can be completed with anonymised input, use that version; if uncertainty remains, escalate to the team that owns data policy. That gives staff a repeatable judgement, not just a prohibition.
What governance and nudges keep training effective
Training alone rarely holds the line unless governance makes the safe behaviour easy. Organisations should back the lesson with clear acceptable-use guidance, approved AI tool lists, and visible reminders at the point of use. Just-in-time nudges work best when they appear where the risky choice happens, such as in browser prompts, upload warnings, or workflow banners.
The strongest programmes also reinforce training with monitoring and feedback. If employees repeatedly try to paste sensitive content into public AI tools, that is usually a signal that the rule is unclear, the approved alternative is too awkward, or the business pressure to move fast is overwhelming the control. The remedy is to adjust the workflow, not just to repeat the warning.
For a broader policy backbone, AI Acceptable Use Policy Template helps translate training into concrete rules for approved tools, data classes, and employee acknowledgement. Organisations should also look at Generative AI Data Loss Prevention Guide for controls such as prompt redaction, browser controls, and logging boundaries that make the training stick in daily use.
Risk and Threat Considerations
Public AI tools create leakage risk because users can paste material that is confidential, proprietary, regulated, or security-sensitive, and that content may then be retained, logged, or exposed through later misuse. The main danger is not malicious intent alone, it is normal productivity behaviour crossing a trust boundary the employee does not fully see.
Failure mechanism: A worker treats the tool like an internal assistant, submits raw data, and the organisation loses control over where that data is stored, reviewed, or reused. If credentials, incident details, customer data, or source code are involved, the same mistake can also expand compromise impact beyond the original prompt.
Impact: The result can be data leakage, contractual or regulatory exposure, loss of intellectual property, and follow-on security incidents if secrets or operational details are disclosed. In practice, the blast radius grows quickly when the same public tool is used by many staff in many roles.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Non-Human Identity Top 10 and OWASP Agentic AI Top 10 address the attack surface, NIST CSF 2.0 and CIS Controls v8 set the technical controls, and ISO/IEC 27001:2022 defines the regulatory obligations.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.AA-05 — Identity Management, Authentication and Access Control | Training must limit who can use approved AI tools and what data they can handle. |
| Recommendation — Define access rules for AI tools and restrict sensitive data entry by role. | ||
| CIS Controls v8 | CIS-3 — Data Protection | The subject is reducing sensitive data exposure to public AI tools through handling rules and DLP. |
| Recommendation — Classify data and apply DLP checks before users can paste sensitive content into AI tools. | ||
| ISO/IEC 27001:2022 | A.5.10 — Acceptable use of information and other associated assets | Employee AI training depends on acceptable-use rules for information handling in external services. |
| Recommendation — Publish acceptable-use rules that explicitly govern what staff may enter into public AI tools. | ||
| OWASP Non-Human Identity Top 10 | NHI-02 — Secret Leakage | Public AI prompts can expose credentials and tokens if staff paste them into chat tools. |
| Recommendation — Train staff to recognise and remove secrets before any prompt reaches a public AI service. | ||
| OWASP Agentic AI Top 10 | ASI09 — Human-Agent Trust Exploitation | Users can overtrust AI tools and share data they would not disclose elsewhere. |
| Recommendation — Use scenario training to teach employees where AI output and data handling can be trusted. | ||
Practitioner Guidance
What to prioritise: Train for classification judgement first, not tool enthusiasm. Employees should be able to decide, in seconds, whether the input is safe to share externally, safe only after redaction, or prohibited entirely.
What to verify: Check that the training includes role-specific examples, a clear escalation path, and an approved alternative for common business tasks. If staff are forced to choose between productivity and compliance, the policy will be ignored in practice.
What good looks like: Teams use public AI for low-risk drafting and ideation, but switch to approved environments or sanitised inputs when the task touches sensitive information. The organisation should see fewer ad hoc exceptions, fewer prompt-related incidents, and more consistent employee decisions.
Practitioner takeaway: The objective is not to stop employees using AI, it is to make the safe choice the easy, rehearsed choice at the moment of use.
Related resources from NHI Mgmt Group
- Sensitive Data
- How should organisations prepare enterprise data for AI use without exposing sensitive information to public LLMs?
- What breaks when employees use AI tools inside browser sessions without data controls?
- How should security teams control shadow AI use when employees paste sensitive data into public models?
Deepen Your Knowledge
Free weekly newsletter
Subscribe to the NHI & AI Identity Journal
The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.
Bonus 33% off our NHI Course when you subscribe.
Reviewed and updated by the NHIMG editorial team on October 30, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org