You trust us with
everything that matters.

This page explains exactly what happens to your data — not in policy language, but in plain terms. What is stored. What is discarded. What you can do. What we will never do.

🔒 AES-256 encryption at rest
🚫 Zero data sold. Ever.
📄 Documents discarded after processing
✂️ Delete everything anytime
"We never sell. Not your data.
Not our values. Not the company you trust."

Vithropic founding commitment

Four practices.
No exceptions.

1

Encryption everywhere

All data stored in Vithropic — your transactions, balances, health metrics, briefing history, and account details — is encrypted at rest using AES-256 encryption. All data in transit uses TLS 1.3. No data moves in plaintext. The encryption applies to everything we retain, and what we don't retain never reaches storage in the first place.

AES-256 at rest TLS 1.3 in transit Zero plaintext movement
2

Secrets management — no credentials in code

API credentials, database connection strings, and service keys never exist in code or configuration files. All secrets live in a dedicated, access-controlled vault with full audit logging — every access is recorded, timestamped, and reviewed. Our application services authenticate using short-lived managed identity tokens. There are no long-lived credentials that can be compromised, because they don't exist.

Zero credentials in code Full audit trail Managed identity auth Azure Key Vault
3

Principle of least privilege

Every service in Vithropic's infrastructure has access only to what it needs to do its specific job — nothing more. The service that processes your uploaded documents cannot access your health data. The service that generates your briefing cannot modify your account. Access boundaries are enforced at the infrastructure level, not just in application code, so a compromise of any single service cannot cascade into the rest of the system.

Role-based access control Service isolation Infrastructure-level enforcement
4

Continuous security testing

Automated vulnerability scans run quarterly against all Vithropic systems. Full external penetration tests are conducted by independent third-party security firms twice per year — using the same techniques real attackers use. Findings are classified by severity and remediated on strict timelines. Critical issues within 24 hours. High severity within 7 days. Results are published in our annual transparency report. Security is not a launch checklist. It is an ongoing practice.

Quarterly automated scans Semi-annual pen tests 24hr critical response Published results

What happens when
you upload a document. Exactly.

Most tools are vague about what they do with your data. This is exactly what happens, step by step, from the moment you upload a statement to the moment it's gone.

📤

Upload and transit

Your document is uploaded over an encrypted TLS 1.3 connection. It is never transmitted unencrypted. The upload goes directly to our processing service — it does not touch permanent storage at any point during transit.

⚙️

Processing in memory

The document is processed entirely in memory. Claude reads the text and extracts transactions and balances. No intermediate file is written to disk. No copy is created. The original document exists only in RAM during this step.

🗑️

Discarded immediately

The moment processing is complete, the document is gone. Not archived. Not backed up. Not logged. The memory is released. What remains is only the intelligence the document generated — transactions and balances.

✓ What we store
  • Transaction amounts, dates, and merchant categories
  • Account balances and closing totals
  • Institution type and account type (e.g., checking, 401k)
  • Statement period dates
✕ What we never store
  • Your name or address from the document
  • Account numbers — full or partial
  • Routing numbers or bank identifiers
  • The document itself — in any form
  • Any image, scan, or copy of the original

What the AI sees
when it reads your document.

Vithropic uses an AI extraction layer to parse transactions and balances from uploaded documents. Here is exactly what that means for your data.

What the AI receives

  • The raw text content of your uploaded PDF
  • A scoped instruction to extract transactions and balances only
  • No prior context about you — every document is processed fresh with no memory of previous uploads

What the AI is instructed to ignore

  • Your name, address, or any personal identifiers on the document
  • Account numbers — full or partial
  • Anything beyond transaction amounts, dates, categories, and balances

AI infrastructure standards

Our AI processing layer operates under enterprise-grade data handling terms — document content is not used for model training and is not retained after processing completes. The extraction call is stateless: nothing from your document persists beyond the transaction and balance data returned to Vithropic's storage layer.

You own your data.
We mean that literally.

These are not aspirational statements. They are specific things you can do at any time, from your account settings, without contacting us.

📤

Export everything

Request a complete export of all your stored data — transactions, health metrics, briefing history, and check-in responses — in standard machine-readable formats. One click. Delivered to your registered email address within 24 hours.

🗑️

Delete everything

Request full account deletion. All your personal data is removed within 30 days. Confirmation is sent when deletion is complete. No questions, no friction, no retention period. The data is gone.

🔍

See exactly what we hold

View a complete inventory of every data category Vithropic holds about you — what it is, how it's used, and when it was last updated. No hidden data categories. No surprises.

🚪

Stop uploads anytime

Stop uploading at any time. Previously uploaded data remains in your account until you choose to delete it. No future data is collected unless you upload again. You control the cadence entirely.

ℹ️
One honest disclosure on deletion

Any anonymized data already contributed to aggregate benchmarks — the statistical patterns used to show how similar profiles compare — cannot be individually extracted after the fact. It was anonymized before contribution and is no longer distinguishable as your data within the statistical population. All personally identifying information is deleted completely. This is disclosed because you deserve to know it.

How peer benchmarks work.
And what they never contain.

What gets aggregated

  • Anonymized behavioral patterns — spending categories, savings rates, sleep averages
  • Outcome data — what configurations produce better results over time
  • Cohort statistics — ranges for similar age, income, and family structure profiles

What is never aggregated

  • Your name, email, or any identifying information
  • Your exact account balances or transaction amounts
  • Your specific health metrics or biomarker values
  • Any data that could identify you within a statistical population

The minimum cohort protection

Cohort benchmarks are only published when a minimum of 50 similar profiles contribute to the statistical pool. Below that threshold, no benchmark is shown — because below that threshold, individual reverse-identification becomes theoretically possible. We treat that boundary as a hard limit, not a guideline.

Three things we will
never do.

These were decided before the first customer joined. They cannot be changed by any offer or acquisition pressure that arrives later.

1
We will never sell your data

Not a privacy policy clause. A founding commitment. Vithropic's revenue comes from subscriptions. Your data is not the product. This is the reason we built a subscription model instead of an advertising one.

2
We will not sell the company for the data

Acquisition offers will come — not for the product, but for the longitudinal behavioral data underneath it. The answer is no. The data was entrusted to us by the people who generated it. It belongs to them.

3
We will not change these commitments

Terms of service can be updated. Privacy policies can drift. These commitments are public, dated, and documented — not buried in legalese. Changing them would require publicly reversing a named commitment.

Where we are.
Where we're going.

Security is not a destination. We publish this roadmap because transparency about what isn't done yet is part of doing it honestly.

Launch — complete

AES-256 encryption at rest, TLS 1.3 in transit, Azure Key Vault secrets management, managed identity authentication, role-based access control, MFA on all admin accounts, document discard-after-processing architecture

Month 1 — in progress

JWT validation on all function endpoints, input validation hardening, rate limiting on all public surfaces, JIT VM access (no persistent RDP)

Month 3

Full API security review, automated dependency vulnerability scanning in CI/CD pipeline, penetration test commissioned

Month 6

First third-party penetration test completed. Findings reviewed, addressed, and published in initial transparency report

Year 1

SOC 2 Type II audit process begins. Annual transparency report published. Semi-annual penetration testing schedule locked

Year 2

Dedicated security role hired. Public bug bounty program launched. Continuous automated security monitoring in production

Found something? Tell us first.

If you believe you've found a security vulnerability in Vithropic, please contact us before disclosing it publicly. We take every report seriously and respond within 24 hours — critical findings within the hour. Responsible disclosure will always be acknowledged and credited.

Report a vulnerability →