The problem
- PHI sent to OpenAI.HIPAA violation on day one.
- Cloud logs your data.OCR fines up to $1.9M/year.
- Vendor promises don't hold.One breach = reputational collapse.
The solution
PHI stays on the edge. Always.
PHI detected at the Cloudflare edge and tokenized before any cloud API call. The model receives anonymized tokens — not patient names, diagnoses, or SSNs. An architectural guarantee you can demonstrate to OCR auditors, not a contractual promise.
HIPAA § 164.312(a)
"Covered entities must implement technical security measures to guard against unauthorized access to ePHI transmitted over electronic communications networks."
PHI stripped before the API call. The model reasons about the case — not the patient.
Frequently asked questions
Does a signed BAA with OpenAI make my app HIPAA compliant?
No. A BAA is a contract — it does not stop PHI from being processed and logged in a third party’s systems, and HIPAA still requires technical safeguards (§164.312). Privedge removes the exposure: PHI is tokenized at the edge before any cloud call, so the model never receives patient data.
If PHI never reaches OpenAI, do I still need a BAA with them?
For the data Privedge tokenizes, OpenAI never processes PHI, so it is not acting as a business associate for that data. Review your full data flows with counsel — a BAA remains good practice for any residual PHI path.
Which patient identifiers does Privedge detect?
It targets the HIPAA Safe Harbor identifiers (§164.514(b)) — names, dates, MRNs, SSNs, contact details and more — using regex plus optional NER. Detection is configurable, so you can add rules for your own record formats.
Can I prove this to an OCR auditor?
Yes. Every request returns pii_matches and routed_to, and only metadata is retained (never prompt content). That gives you a verifiable, request-level trail showing PHI was tokenized before it left your network.