Audit AI privacy budget integrity

What it's for

Replaces privacy checkboxes with an auditable privacy-spend ledger: leadership can see exactly which AI releases consumed each population's formal budget and whether the claimed bound actually composes.

What you give it

Inputs split into evidence read from your connected systems, calibration your team owns, and numerical controls that affect precision but never the result's meaning.

{
  "privacy_accounts": [
    {
      "claimed_composed_epsilon": 1.5675283643313485,
      "evidence_verified": true,
      "id": "customer-analytics-account",
      "maximum_epsilon": 2,
      "period_ended_at_ms": 10000,
      "period_started_at_ms": 0,
      "permitted_purpose_ids": [
        "support-analytics"
      ],
      "privacy_population_id": "eu-customers",
      "rdp_orders": [
        2,
        4,
        8,
        16
      ],
      "target_delta": 0.00001
    }
  ],
  "privacy_releases": [
    {
      "evidence_verified": true,
      "id": "support-statistics-release-1",
      "mechanism_type": "subsampled_gaussian",
      "privacy_account_id": "customer-analytics-account",
      "privacy_review_approved": true,
      "purpose_id": "support-analytics",
      "rdp_epsilon_by_order": [
        0.1,
        0.2,
        0.4,
        0.8
      ],
      "release_hash": "sha256:support-statistics-release-1",
      "release_sequence": 1,
      "released_at_ms": 1000,
      "sampling_probability": 0.1,
      "subject_count": 10000
    }
  ]
}

What you get back

This is the actual output of running the example above — computed by the same function the platform calls, not an illustration.

{
  "account_diagnostics": [
    {
      "best_rdp_order": 16,
      "claimed_epsilon": 1.5675,
      "composed_epsilon": 1.5675,
      "epsilon_candidates_by_order": [
        11.6129,
        4.0376,
        2.0447,
        1.5675
      ],
      "epsilon_reconciliation_gap": 0,
      "failed_gates": [],
      "maximum_epsilon": 2,
      "privacy_account_id": "customer-analytics-account",
      "privacy_population_id": "eu-customers",
      "rdp_epsilon_by_order": [
        0.1,
        0.2,
        0.4,
        0.8
      ],
      "release_count": 1,
      "target_delta": 0
    }
  ],
  "assumptions": [
    "Every RDP vector already includes the approved mechanism's sampling-amplification analysis and uses the exact common order grid declared by its privacy account.",
    "A privacy account represents one governed neighboring-population relation; releases that can touch the same person are never split across accounts to evade composition.",
    "RDP composes additively and converts to the represented approximate-DP epsilon at the target delta; the accountant does not infer mechanism parameters from labels."
  ],
  "counts": {
    "budget_breaches": 0,
    "invalid_releases": 0,
    "privacy_accounts": 1,
    "privacy_releases": 1,
    "reconciled_accounts": 1
  },
  "decision": "privacy_budget_ledger_verified",
  "evidence": {
    "gate_pass": true,
    "minimum_required_fraction": 0.95,
    "verified_fraction": 1

How it works

Statistical audit & measurement — Check whether a number is fit to decide on: coverage, timing, reconciliation, and the gaps a dashboard hides.

1. 1 Validate each privacy account's neighboring population, strictly increasing RDP order grid, target delta, budget, purpose perimeter and accounting period.
2. 2 Join every immutable private release to one account and audit consecutive sequence, unique hash, time, purpose, mechanism, review and aligned precomputed RDP vector.
3. 3 Sum RDP at every order, convert to the tightest represented approximate-DP epsilon, reconcile the claim and freeze releases when evidence or budget gates fail.

Before you trust it

Every tool in the catalog ships with the conditions under which its answer is meaningful — and the conditions under which it should abstain instead of guessing.

Calibrating it to your org

Calibration workflow

1. 1 Define the management decision, target outcome, aggregate unit, privacy boundary, cadence, and prediction/intervention horizon for this organization.
2. 2 Build a tenant-scoped historical cohort using only information available before each prediction or decision; preserve zero periods, censoring, assignment probabilities, and unresolved outcomes when the method requires them.
3. 3 Estimate statistical parameters on training history, but obtain costs, utilities, risk tolerance, practical-effect thresholds, capacity, and policy constraints from accountable decision owners.
4. 4 Validate on later time windows or held-out aggregate units at the deployment grain, against a simple baseline and the function-specific validation strategy.
5. 5 Deploy only if the returned decision clears evidence, overlap, calibration, robustness, and guardrail checks; warning, unsupported, schema-gap, and fallback decisions are abstentions.
6. 6 Monitor realized outcomes, data drift, coverage, and decision regret; recalibrate at a governed cadence or after a detected regime/definition change, never merely because a stakeholder dislikes the result.

Call it from your AI

You don't wire up 388 tools in your MCP client. The GitRevio MCP server exposes 18 tools, three of which let an agent search the catalog, read a tool's schema, and run it — so the assistant finds this one on its own.

gitrevio_capabilities_search
  { "q": "recompute each aggregate ai privacy account" }
  → finds "audit_ai_privacy_budget_integrity"

gitrevio_capability_describe
  { "capability_id": "audit_ai_privacy_budget_integrity" }
  → returns the input schema and agent guidance shown on this page

gitrevio_capability_run
  { "capability_id": "audit_ai_privacy_budget_integrity", "arguments": { ... } }
  → returns the result shown above

Works in Claude Desktop, Claude Code, Cursor, Cline, Continue.dev, Goose and Aider. See the MCP server.

Related tools

See every tool in AI risk, rights & assurance →