01 · unsafe pasteRaw stays local
“Draft the appeal for patient Delia Marchetti, MRN 4471902, plan ID BCX99188342.”
Customer-hosted Sector policy Chain verified
When they ask how your team used AI, you already have the file.
One customer-owned control point between sensitive work and an approved model. Policy runs first; operating evidence stays with you.
raw prompt stays local
policy v1.0 enforced
evidence ready
Accountingtax IDs, payment rails, client figures
LedgerProof HealthMRNs, plan IDs, dates, contact data
Risk & compliancecustomer, network, and control evidence
Customer-ownedpriced by environment, not by seat
The sample pack
See the proof before the pitch.
This is a real Evidence Pack generated by Ledgerproof for a fictional six-person firm over the 2026 filing season. Synthetic data, real product, nothing gated.
Chain verified
Harbor and Vance CPAsSample Evidence Pack · synthetic data
Harbor and Vance CPAsSample Evidence Pack · synthetic data
What is in the file
- Cover
Firm, period, policy versions, chain status.
- Chain attestation
Record count, first and last hash, verification result.
- What was used
Interactions by user and backend, week by week.
- What was redacted
Counts for SSN, EIN, account, client name, and trial balance figures.
- What was blocked
Every block event with the policy version and staff user.
- Reviewer sign-off
A documented review of AI output.
Honest scope: the pack attests to what flowed through the gateway. It does not claim to see what never touched it.
The live control
See what survives before the model does.
Try a synthetic sector example or write your own. This illustrative pass runs only in your browser; nothing is sent anywhere.
Raw promptstays local
What the model would seeblocked
Safe retry draftgenerated locally · never auto-sent
LedgerProof Health
Patient identifiers stop here. The useful work can continue.
Start with one administrative text workflow—follow-up instructions, prior-authorization language, or an internal policy draft. Direct identifiers can block the request; lower-risk classes can become typed tokens before an approved model sees the prompt.
Health
Customer-owned AI control & evidence
Scope a Health pilot
Policy-controlled masking is not, by itself, a HIPAA de-identification determination or compliance certification. The customer’s privacy, security, and legal owners remain in charge.
02 · decisionBLOCK
MRN and plan ID are forbidden classes. Nothing reaches the model; the event is still recorded.
03 · guided retryHuman reviewed
“Draft the appeal for PATIENT_NAME_1. Keep the rationale administrative and route it for clinician review.”
First workflowAdministrative drafting, text only
Quality gate30 synthetic cases, 6 held back
Hard boundaryNo EHR, diagnosis, or autonomous action
Your staff did not wait for your AI policy.
Every unapproved prompt can carry a Social Security number, an EIN, or a client ledger. Right now, most firms have no durable record of what left, when, or to whom.
What gets installed
A control, not a binder.
Detection, policy, redaction, backend choice, and the evidence chain. One owned system in your environment.
Swipe to inspect the installed stack →Gateway
One internal endpoint staff use instead of raw chatbots. Every request passes through the same rules.
FastAPI · local or private cloudPolicy
Allow, redact, or block per data class in a YAML file a control owner can read.
Fail-closed defaultsMCP
The same control inside Claude Desktop through a governed tool with the chain already baked in.
Installed in minutes
D
Owned
Your box, your keys, your policy, your log.
Console
The actual operator surface, not a design mockup.
The engineered peak
Change one number. Watch the proof fail.
Each record carries the hash of the record before it. Alter one event and every proof downstream breaks.
Swipe the records → record 03 is editable.
5fecd5c9320c4e4f81148dde28d69e2f$ ledgerproof verify audit.jsonl
CHAIN VERIFIED (5 RECORDS)
Sealed. Every hash matches the record and its predecessor.
The evidence map
Every page answers a duty you already have.
Read across: the pack section, the control it demonstrates, and the question it helps answer. Your counsel confirms the mapping for your firm.
Pack sectionControl evidencedWhat the reviewer can verify
Policy versionsWritten, versioned security program
Which rules were in force for each interaction.
What was usedAI-channel risk inventory
Who used which approved backend and when.
What was redactedData minimization in operation
Which client-data classes were removed before forwarding.
What was blockedPolicy enforced, not merely written
Every exception with user, decision, and policy version.
Chain attestationMonitoring and control testing
Whether the operating record has been altered.
Reviewer sign-offDocumented review of AI output
Who reviewed the output, and when.
Sources: 16 CFR Part 314; IRS Publication 4557; IRS OPR guidance 2026-19. This is an engineering mapping, not legal advice.
Show me where your WISP says it. Then show me you are doing it.
The second sentence is where paper-only programs fail. Ledgerproof supplies the operating record, while keeping the boundary honest.
Ledgerproof producesApproved endpoint usage, chain verification, AI-output review, quarterly Evidence Pack.
Shared responsibilityAI risk assessment, written policy, block-event response.
Your firm suppliesThe full WISP, training acknowledgements, incident plan, and vendor agreements.
Owned outcomes
Three clear ways to begin.
You buy a working control and its evidence, not seats on a meter that reprices every year.
Swipe to compare the three starting points →Map before you move
Shadow-AI Control Sprint
$2,500fixed · credited toward an install within 60 daysExposure map, sector-aware policy draft, synthetic acceptance tests, and a first Proof Folder.
Find my starting pointControlled teams
Governed AI Install
$6,500fixed · single environment · 15-day deliveryOne approved workflow live, staff onboarded, tested policy, operating runbook, and the first verified Evidence Pack.
Plan the installStay current
Evidence Assurance
$750/qtrcancel anytime · keep every artifactMonthly chain check, quarterly pack, exception summary, guidance bulletin, and owner review.
Check assurance fitMore than one office or identity domain?
Start with a multi-office design pilot.
Scoped by environments, identity, integrations, and policy tiers—never by seat. Wider rollout begins only after the pilot proves the operating and evidence model.
Scope the pilot →A guided starting point
Three answers. One sensible next step.
Ledgerproof is not only an SSN masker. The policy changes with the data, workflow, and control environment—from tax identifiers and payment rails to health-plan IDs, medical record numbers, network details, and confidential control evidence.
- 01Mapworkflows and sensitive classes
- 02Decideallow, redact, or block
- 03Testwith synthetic acceptance cases
- 04Deployone approved pathway
- 05Proveverify, pack, and improve
What you receive in 15 days
A controlled pilot, not a compliance promise.
- Current-state exposure map
- Readable sector policy
- Customer-relevant synthetic tests
- One approved AI pathway
- Profile preflight and synthetic probe
- First verified Evidence Pack
- Runbook, limits, and next actions
Built and installed by one accountable person
The system comes with the engineer.
Umair Tareen brings twelve years in enterprise data, including a decade across banks and six years at MUFG. Those systems lived under audit. Ledgerproof's habits come from that environment: redaction pipelines, append-only logs, and evidence that survives hostile review.
He installs Ledgerproof personally, in your environment, and you keep what he builds.
Straight answers.
Does sensitive data pass through LedgerProof servers?
No. There are no LedgerProof servers in the data path. The gateway runs on customer-controlled infrastructure, and model keys stay there.
Which AI backends does it support?
OpenAI, Anthropic, or fully local through Ollama. Your keys, your choice, configurable per install. The public demo uses a deterministic local stub.
What stops staff from using ChatGPT on a phone?
Nothing technical, and anyone claiming otherwise is selling surveillance. The install pairs a governed path with a signed policy and staff onboarding. The Evidence Pack states exactly what the control covers.
We already have Microsoft Purview, Copilot, or model guardrails. Why this?
Those are real controls and may be the better fit for their native environments. Ledgerproof is compelling when you want a smaller self-hosted, model-neutral boundary with a readable policy, local metadata-only audit trail, and portable evidence you can verify offline. It complements an enterprise security stack rather than pretending to replace one.
Does it protect more than Social Security numbers?
Yes. The shipped profiles cover accounting and payment data plus text-based healthcare and bank-risk identifiers such as EINs, accounts, names, dates, MRNs, NPIs, plan IDs, addresses, email, phone, IPs, device IDs, licences, vehicles, and biometric references. Detection is contextual and must be tested against each customer's synthetic examples before live use.
Is this priced per seat?
No. The install is scoped by environment, workflow, backend, and support boundary. The customer owns the gateway, policy, log, and every Evidence Pack. Quarterly assurance is optional.
Is this legal advice?
No. Ledgerproof is an engineering control mapped to professional duties. Your counsel and your judgment stay in charge.
A 15-minute walkthrough
Use a synthetic example. Never send sensitive data here.
Tell Umair the team size, sector, and first workflow. You will get two time options within two business days.
Sample data onlyNo obligationTwo-day reply