Security
A tool that checks access shouldn't need any.
How VerifAIdentity is built to be trusted with identity data: what flows where, how companies are kept apart, how credentials are protected, how evidence stays verifiable, and where it can run.
What flows where
Data comes in through read-only connectors and goes out only to destinations you configure. There is no path back into your identity systems.
Keeping companies apart
Two layers: separate infrastructure for each company in pilots, and separation enforced in the code for when instances are shared later.
- A dedicated instance per company
- In the hosted design each design partner gets its own resource group, container environment, PostgreSQL database and key vault. Deleting a pilot deletes the resource group.
- Every row carries its company
- All 18 data tables carry a company ID, and every query passes through one layer that adds it. No page or rule can forget it.
- Tested at four levels
- Two companies with identical IDs and usernames in one store: the schema, every read (including the audit pack, SIEM payload and brief), every write, and the evidence chain show no leakage either way. Run on SQLite and PostgreSQL.
- The assistant sees one company
- Each answer is built only from the asking company's records; there is no shared context or memory.
Credentials and data, handled separately
- Read-only permissions
- Every connector asks for read permissions only, listed on the Integrations page. The Entra and PingOne connectors check their own granted roles and warn if anything can write.
- Secrets in a vault, never in the product data
- Connector secrets and the evidence signing key live in Azure Key Vault in the hosted design, or in a protected file (mode 600) in the lab. They never enter the database, logs, exports or screenshots.
- Your credentials can stay with you
- In the hybrid option the connectors run inside your network and send results out over HTTPS; our cloud never receives your API credentials.
- Only the identity data the rules need
- Names, usernames, IDs, status, roles, approvals, HR dates and sign-in events. Never passwords or hashes, MFA secrets, emails, messages, files, or salary and health data.
- App secrets: dates only
- For app registrations we keep each secret's ID, name and start and end dates. Microsoft Graph never returns secret values, and we drop hints.
- Encrypted in transit and at rest
- TLS for every connection; managed PostgreSQL encrypts data at rest.
Evidence that stays verifiable
A fingerprint stored next to a record proves nothing by itself: whoever can edit the record can recompute it. So every capture is also appended to a chain.
- Each entry stores the record's SHA-256, links to the previous entry's hash, and is signed with HMAC-SHA256 using a key outside the database
- Detected: a record edited after capture, an entry edited, inserted, deleted or reordered, and a chain rewritten without the key
- Not detected by the chain alone: someone holding the key who rewrites everything, or removes the newest entries. Keeping the chain head outside VerifAIdentity (SIEM, WORM storage, the audit file) closes that: any later check against it fails
- Planned: trusted timestamps (RFC 3161) on the chain head
Method EC-1.0. Covered by ten automated tests, on SQLite and PostgreSQL.
Where it can run
Today the product runs in our lab. These are the designed options; none is deployed for a customer yet.
| Topic | Dedicated hosted instance | Hybrid: collector in your network | Private: in your cloud |
|---|---|---|---|
| Status | Designed, not deployed | Designed, not deployed | Designed, not deployed |
| Runs on | Azure Container Apps, Azure Database for PostgreSQL (private access), Key Vault, one resource group per company | Dashboard as hosted; connectors in a container you run | The same container images in your Azure, AWS or Google Cloud account |
| Your API credentials | In a key vault dedicated to your company | Never leave your network | In your own vault |
| Data location | An Azure region agreed with you | Results in the agreed region; raw vendor data stays with you by default | Wherever you run it |
| Sign-in to the dashboard | Microsoft Entra ID, named users only, MFA | Same | Your identity provider |
| Backups | Automatic database backups; a restore is tested before the first pilot | Same | Your backup policy |
| Upgrades | We deploy tagged container images and tell you in advance | You choose when to update the signed collector image | You choose when to deploy new images |
| Who operates it | VerifAIdentity | Shared: you run the collector, we run the dashboard | You, with our support under agreed access |
| Network | Database and vault on private endpoints; outbound only to identity vendor APIs | Outbound HTTPS from the collector only; no inbound ports | Your network rules |
Where we are, plainly
VerifAIdentity is built and tested on lab data in our own tenants. It has not been independently audited or penetration tested, and we do not have a SOC 2 report. Both are planned when a customer contract funds them.
Before a pilot we share the architecture, the data flow, the exact connector permissions and our test results, and we answer your security questionnaire.
Report a security issue
If you find a vulnerability in this website or the product, email venkata@verifaidentity.com with the subject "Security report". We aim to reply within two working days and will keep you updated until it is fixed.
Please don't access data that isn't yours, run denial-of-service tests, or test against anyone else's account. Our security.txt has the same details.