# Data minimization — only necessary context sent to AI

- **Pattern:** `ab-000162` (`ai-data-privacy.pii-protection.data-minimization-in-prompts`)
- **Severity:** medium
- **Lifecycle:** active
- **Last modified:** 2026-04-18
- **Canonical URL:** https://auditbuffet.com/patterns/ab-000162
- **License:** CC-BY-4.0 — attribute to AuditBuffet Pattern Catalog (https://auditbuffet.com/patterns/ab-000162)

## Why it matters

GDPR Article 5(1)(c) codifies the data minimization principle: personal data must be adequate, relevant, and limited to what is necessary for the processing purpose. CCPA §1798.100 establishes parallel minimization expectations. When a developer serializes a full user object into an AI prompt — including email, hashed password, billing address, internal IDs, and account metadata — fields that have no bearing on the AI's task travel to the provider. The AI does not need the user's email to write a support reply. It does not need their `created_at` timestamp to summarize a document. Every unnecessary field is data shared without purpose, creating regulatory exposure and expanding the blast radius of any future provider breach.

## Severity rationale

Medium because full-object serialization into AI prompts sends unnecessary personal data to a third-party processor in violation of GDPR Art. 5(1)(c), but the harm requires a downstream event (breach, subpoena) to materialize.

## Remediation

Build an explicit, minimal prompt context object rather than serializing the full user or database record. Document which fields each AI feature actually requires.

```ts
// Instead of:
const context = JSON.stringify(user) // Sends email, hash, billing, internal IDs

// Use:
const context = JSON.stringify({
  name: user.firstName,
  plan: user.subscriptionTier,
  // Only fields the AI prompt template actually references
})
```

Create a dedicated `buildAiContext(user: User): AiUserContext` function in `lib/ai/context.ts` that exports a typed, narrowed object. This makes it easy to audit what is being sent and prevents accidental field additions when the `User` model grows. Review the exported fields whenever new columns are added to the user record.

## Detection

- **ID:** `data-minimization-in-prompts`
- **Severity:** `medium`
- **What to look for:** Enumerate every relevant item. Examine how user context is injected into AI prompts. Look for patterns where full user objects are serialized into the prompt (e.g., `JSON.stringify(user)`, `` `User info: ${JSON.stringify(userData)}` ``, spreading a full database row into prompt context). Also look for patterns that construct a minimal context object with only required fields.
- **Pass criteria:** At least 1 of the following conditions is met. When user context is included in prompts, only the necessary fields are selected and injected — not the full user record. Code constructs an explicit subset object before stringifying it into the prompt.
- **Fail criteria:** Full user objects, full database records, or large context objects are serialized directly into the AI prompt, potentially including fields like `email`, `hashed_password`, `billing_address`, `created_at`, or internal IDs that the AI does not need.
- **Skip (N/A) when:** No structured user context is injected into prompts — prompts consist only of the user's direct message without additional context objects.
- **Detail on fail:** `"Full user/data object serialized into AI prompt in [file] — potentially sending fields like email, internal IDs, or other unnecessary data to the AI provider"`
- **Remediation:** The principle of data minimization means sending only what the AI needs to answer the question. A user's name and subscription plan is relevant context for a customer support AI. Their email address, password hash, and billing details are not.

  Build an explicit prompt context object:

  ```ts
  // Instead of:
  const context = JSON.stringify(user) // Sends everything

  // Use:
  const context = JSON.stringify({
    name: user.firstName,
    plan: user.subscriptionTier,
    accountAge: user.createdAt ? 'established' : 'new'
    // Only fields the AI prompt actually uses
  })
  ```

  Document which fields each AI feature requires in a comment or schema. Review this list when adding new fields to user records.

  For a broader review of what data your analytics pipeline captures, the Analytics & Tracking Audit covers data minimization in event tracking.

## External references

- gdpr Art. 5(1)(c)
- ccpa §1798.100
- nist-ai-rmf MAP 1.5
- cwe CWE-359

Taxons: privacy-consent, inference-contract

HTML version: https://auditbuffet.com/patterns/ab-000162