HIPAA & Security

Our Safe Harbor Architecture

We follow the HIPAA Safe Harbor de-identification method by removing the 18 identifiers before any clinical text or file is processed by our backend. Any data that is stored in your ChiroReports account is de-identified by default.

• Browser-side redaction: When you upload an imaging report PDF to the X-Ray & MRI Explainer, the PDF is opened and read in your browser. Names, dates of birth, medical record numbers, dates, phone numbers, and SSNs are removed and replaced with tokens like [PATIENT] and [DOB] before the content is sent to our servers.
• No patient-identifiable report list: We do not store the original uploaded file or image, and the dashboard does not maintain a Recent Reports section. If draft content is saved, it is de-identified and uses placeholders instead of patient identifiers.
• Patient identity stays on your device: The patient name and date of birth you enter into the local Identity panel are stored only in your browser's localStorage. They are never sent to our backend. The real name is merged into the report at the last mile, in your browser, for preview, printing, and export.
• Encrypted in transit and at rest: All traffic between your browser and ChiroReports uses TLS. Any de-identified account data is stored in Lovable Cloud and protected by the same encryption standards.

What This Means for You

Because PHI is removed before it reaches our infrastructure, we do not act as a business associate with access to your patient records. You remain in control of the real patient identity at all times, on your own device.

This approach does not replace legal advice, and you should still use your practice's own HIPAA policies for the devices and accounts your staff use. If you have questions about this architecture, contact us at support@chiroreports.app.

AI Processing

We use AI to translate clinical language into plain English and to generate report text. Only de-identified text or redacted image data is sent to the AI gateway. We do not use your uploaded reports to train models, and we do not retain prompt data longer than necessary for the request.