How do you prevent sensitive data from being included in AI prompts?

How to answer this security questionnaire question, with an expert response your security or GRC team can adapt.

Expert answer

Apply input filtering, redaction, and data-classification controls before content reaches a prompt, and restrict which data classes may be sent to AI systems at all. Combine automated detection of secrets and regulated data with clear policy and developer guardrails.

What the security reviewer is checking

This question tests whether your controls are technical or merely aspirational. A policy telling employees not to paste secrets into AI tools does not satisfy a reviewer on its own; they want to see enforcement layers — data classification rules wired into the pipeline, automated detection and redaction of credentials and regulated identifiers before content reaches a model, and restrictions on which data stores an AI feature can even read from. For product AI features specifically, reviewers ask what flows into prompts by architecture, not by accident, and whether provider agreements prohibit retention and training on that content.

Example response you can adapt

This is an illustrative template, not a real vendor's security posture. Replace every claim with what is actually true for your organization before submitting it.

We control prompt contents through architecture first and filtering second. AI features operate on scoped retrieval: each request can only access data the authenticated user is already authorized to see, drawn from an allow-listed set of content types, so payment data, credentials, and system secrets are never reachable by the prompt-assembly layer. Before any content is sent to a model, an automated pre-processing step scans for and redacts secrets (API keys, tokens, private keys) and high-risk identifiers using detection tooling in the request path. All model calls route through a central gateway that enforces these rules uniformly, logs metadata for audit, and connects only to providers under enterprise terms with zero data retention and a contractual bar on training. For employee use of AI tools, our acceptable use policy limits sensitive data to approved, contracted tools, backed by security training and periodic access reviews.

Evidence reviewers expect you to attach

Data flow diagram for AI features showing what can and cannot enter a prompt
AI acceptable use policy covering employee handling of sensitive data
Configuration or design documentation for the redaction / secret-detection step
Data processing terms with AI providers showing retention and no-training clauses

Follow-up questions reviewers ask next

Is the redaction step enforced in code for every model call, or applied selectively?
Which categories of data are explicitly barred from AI processing?
How would you detect it if sensitive data did reach a model provider?
Do prompts or completions get logged, and who can read those logs?
Are employee uses of external AI tools monitored or restricted at the network level?

Answer every security questionnaire in minutes

Wolfia drafts accurate, cited answers to security questionnaires and RFPs from your existing documentation. See it work on your own questions.Book a demo

Related ai & ml questions

Do you use any AI or machine learning in your product?How do you handle customer data used in AI features?Do you use third-party AI APIs (OpenAI, Anthropic, etc.)?What data is sent to AI service providers?Do you have an AI acceptable use policy?Can customers opt out of AI features processing their data?

Browse the full security questionnaire question library