Skip to content
Mental Health · HIPAA

Use ChatGPT and Claude on Your Practice's Notes — Without Disclosing PHI

PortEden replaces patient names, DOBs, diagnoses, and chart attachments with placeholders before your prompt reaches OpenAI or Anthropic. Use any AI for SOAP notes; the model never sees the underlying PHI.

See pricing

Free for solo practitioners · No credit card · Per-request audit log built in

Maps to
HIPAA
GDPR
SOC 2
CCPA
Quebec Law 25
The Risk

What Goes Wrong Without PortEden

You Paste a Session Summary Into ChatGPT to Refine a SOAP Note

That summary holds the patient's name, presenting diagnosis, and clinical detail. The moment OpenAI's API receives it, you've disclosed PHI to a third party with no covered-entity contract, no retention guarantee, and no audit trail you can produce for HHS.

You Use Claude to Triage a Crisis Email From a Patient

The thread holds the patient's identity, their history, and language a chart auditor would treat as sensitive. Claude's context window now holds the most identifying record in your practice — for as long as Anthropic's retention policy says.

Your Intake Form Plugin Sends New-Patient Data to an AI

Names, DOBs, insurance IDs, and presenting concerns flow into a generic AI before they ever hit your EHR. One subprocessor change at the AI vendor and you have a Breach Notification Rule problem you can't unwind.

How PortEden Protects Your Practice

PHI, Redacted Before It Reaches the Model.

PortEden inspects every field your AI is about to see. Patient identifiers, DOBs, diagnoses, and free-text PHI are replaced with placeholders at the boundary — never sent to OpenAI or Anthropic.

Your data
PortEdenRedact
Your AI
Claude
ChatGPT
Copilot
Gemini
Grok
Safe
Sensitive
Redacted
Compliance Reality

What HIPAA Actually Requires When Your Practice Uses ChatGPT or Claude

  • Consumer ChatGPT and Claude tiers do not sit inside HIPAA's covered service-provider chain — pasting PHI into either is a disclosure to a third party that HHS will not treat as authorized.
  • 45 CFR §164.312 requires technical safeguards over electronic PHI: access controls, audit controls, and integrity. "Don't paste PHI into the AI" is not a safeguard a regulator will accept.
  • The HHS Breach Notification Rule treats unauthorized disclosure of unsecured PHI as a breach. AI vendors logging your prompts for training or moderation can land in that bucket.
  • Patients have a right to know how their PHI is used. Op. 512-style transparency is now appearing in state mental-health board guidance — document your AI use the way you document referrals.
HHS HIPAA Privacy Rule
The Solution

Built For Therapists

PHI-Aware Redaction in Gmail and Outlook

Replaces 50+ identifier types — patient names, DOBs, diagnoses, MRNs, insurance IDs, and free-text PHI — with placeholders in under 200 ms before the prompt leaves your perimeter. The AI receives only the redacted version; the original chart never reaches OpenAI or Anthropic.

SOAP-Note Drafting Without the Chart Leaving Your Practice

Paste a raw session summary, get back a structured SOAP note. PortEden redacts the identifying detail before the model sees it and re-hydrates names and dates locally so your final note reads correctly. The AI never sees who the patient is.

Per-Patient Consent and Audit Log

Configure per-patient AI rules at intake. Every prompt that touches a patient's data is logged, timestamped, and exportable as a CSV — the kind of record HHS and state mental-health boards expect when they ask how AI was used in care.

Works With Claude, ChatGPT, and Copilot — No EHR Replacement

Keep your existing EHR, your existing AI tools, and your existing workflow. PortEden sits in front of any model with no plugin to install in SimplePractice, TherapyNotes, or Jane — solo practitioner to group practice in days, not months.

Works With What You Already Use
Gmail
Gmail
Inbox-side redaction for AI replies and triage
Outlook
Outlook
Microsoft 365 inbox-side redaction
Google Calendar
Google Calendar
Strip appointment titles and attendee phone numbers
Teams
Teams
Microsoft 365 chat redaction for care coordination
With and Without PortEden

The Same Workflow, Two Very Different Outcomes

Drafting a SOAP Note With ChatGPT
Without
Patient name, DOB, diagnosis, and clinical detail sent to OpenAI in plain text — a disclosure of PHI to a third party with no covered-entity contract in place.
With
Identifiers and clinical specifics replaced with placeholders before the request leaves your network. The AI structures the note; PortEden re-hydrates locally.
Triaging a Patient Crisis Email With Claude
Without
Full thread — sender, body, attachments — sent to Anthropic. Includes any PHI the patient volunteered, retained per Anthropic's policy.
With
Email reaches the model with patient identifiers and PHI replaced by placeholders. The AI suggests a triage path without seeing who the patient is.
Calendar-Aware AI Assistant on Your Practice Calendar
Without
Appointment titles like "Initial intake — depression screen" and patient phone numbers logged by the AI vendor for the length of their retention policy.
With
Appointment titles and attendee identifiers sanitized before any AI sync. The assistant sees "Patient session" and a redacted slot.
HHS Audit or State Board Inquiry Into Your AI Use
Without
No record of which patient data went to which model. You're reconstructing it from screenshots and the AI vendor's portal — if you still have access.
With
Per-patient, per-session audit log of every prompt, exportable on demand for HHS, your state board, or a malpractice carrier.
Group Practice Rollout to Associates and Interns
Without
Each clinician follows AI policy by hand; one paste-and-prompt is a practice-wide HIPAA problem.
With
Practice-wide redaction defaults; per-patient and per-payer overrides flow from your intake or EHR.
Try It on Your Practice

Five-Minute Setup. Free for Solo Practitioners.

Connect Gmail or Outlook via OAuth. Pick the HIPAA profile. Keep using ChatGPT or Claude exactly the way you do today — with PHI protected by default.

See pricing

Frequently Asked Questions

Does using ChatGPT or Claude with PortEden count as disclosing PHI to a third party under HIPAA?
PortEden replaces PHI with placeholders before the prompt leaves your perimeter. The third-party AI receives only the redacted version, so the underlying protected health information isn't disclosed in the clear. You still owe the usual HIPAA Privacy and Security Rule obligations — minimum necessary, access controls, audit — but the disclosure pathway that worried HHS is closed at the boundary.
How does PortEden help my HIPAA program?
PortEden gives you the technical controls a HIPAA program leans on when AI is in the loop: Safe Harbor-style redaction of the 18 identifier categories at egress, default-deny role-based access controls on what each clinician can prompt against, and a per-request audit log exportable to your SIEM or to a binder for an HHS inquiry. Compliance with HIPAA remains your responsibility — PortEden provides the technical control, you operate the program around it.
Will PortEden change how I write SOAP notes?
No. You still draft the way you draft today, in the EHR or template you already use. PortEden sits in front of ChatGPT, Claude, or Copilot — when you ask the AI to clean up or structure a note, it sees a redacted version, returns a structured note, and PortEden re-hydrates the names and dates on your side before the result lands in your chart.
What happens if a patient asks how I'm using AI in their care?
PortEden's audit log gives you a per-patient record of every prompt that touched their data. A growing number of state mental-health boards now expect transparency about AI use; the same record you'd produce for HHS can be summarized for a patient who asks. We ship a sample consent addendum you can drop into your intake packet.
Does PortEden work with SimplePractice, TherapyNotes, or Jane?
Yes — PortEden runs in front of your AI, not your EHR. You keep using SimplePractice, TherapyNotes, Jane, or whatever EHR you have, and the redaction layer sits between any AI tool you launch and the data that AI is about to read.
Can a group practice apply different redaction rules per clinician or per payer?
Yes. Set practice-wide HIPAA defaults once; per-clinician, per-payer, and per-state overrides flow from your intake or EHR mapping. A telehealth-only Quebec patient can ride a stricter Law-25 profile while a cash-pay client uses the default — no one re-configures anything by hand.
What does it cost and how long does setup take?
There's a free tier for solo licensed practitioners. Group practices scale by clinician — full pricing is on the pricing page. Setup is under 5 minutes for a solo therapist on Gmail + ChatGPT or Claude. Group practices typically take a half-day for SSO and EHR webhooks.

Keep Exploring

Ready to Use AI Without Risking PHI?

Five-minute setup. Free for solo licensed practitioners. HIPAA-aligned audit log from day one — exportable per patient, per session, on demand.

See pricing

Group practice or behavioral-health system? Talk to sales →