By · Last updated 2026-05-29

Back to BlogHealthcare

HHS 2025: AI Clinical Notes Need PHI

AI transcription systems can inadvertently put Patient A's PHI in Patient B's record. Here's why real-time PHI detection before EHR commit is the control.

May 29, 20269 minute read
HIPAA complianceclinical documentationPHI detectionEHR privacyHHS 2025

The AI Clinical Notes Privacy Problem

Updated for 2026

Hospitals and clinics use AI to write clinical notes. AI transcribes voice and drafts text. But this creates a HIPAA gap that manual review cannot close.

AI-generated notes expose patient records in three ways:

  1. Cross-contamination: AI may pull info from one patient into another patient's record. Medical AI studies have shown this risk.
  2. Context bleed: Patient info lands in the wrong field — a billing note, a research field, or a referral form. AI fills fields by context, not by field purpose.
  3. Vendor data use: Many AI vendors send notes back for model review unless you opt out. This sends patient info to third-party servers. Those servers may not have a signed BAA.

HHS published a proposed rule in 2025. It says entities using AI tools must include those tools in their risk analysis. This creates a formal rule for AI-assisted clinical work.

The 2025 HHS AI Risk Analysis Rule

HHS proposed new rules for covered entities that use AI. Each AI system that touches patient records must appear in the entity's risk analysis.

The rule has three parts:

Tech safeguards: Review each AI tool. Ask:

  • Does it send patient records outside your systems?
  • Does it store patient records on its servers after use?
  • Does it write patient info into the wrong record?

Staff training: Training must cover AI-specific risks. This includes record mix-up cases.

Physical controls: Workstations running AI tools must be part of physical access controls.

AI clinical tools include voice-to-text services, AI note drafting tools, and coding tools.

Why Pre-Save Detection Works

The best tech control is PHI detection before the note saves to the EHR.

Without pre-save detection:

  • AI writes the draft
  • Staff reviews it by hand, under time pressure
  • Note saves to EHR
  • PHI errors are now in the permanent record
  • Fixing them requires audit entries and a breach review

With pre-save detection:

  • AI writes the draft
  • PHI scan runs before the note saves
  • Flagged items go to staff for review
  • Staff fixes errors before saving
  • The EHR record is clean from the start

Pre-save detection meets HIPAA Security Rule 164.312(b). That rule requires systems that record and check activity. The pre-save scan creates an audit record for every note reviewed.

The 18 PHI Categories in AI Notes

HIPAA Safe Harbor requires removing 18 categories of PHI (45 CFR 164.514(b)). AI notes can surface all 18 in ways you may not expect:

  • Names — a patient names a family member in symptom history
  • Location — home address in social history
  • Dates — birth dates, admission dates, procedure dates
  • Phone and fax numbers — contact info in referral notes
  • Email addresses — patient-provided contact details
  • SSNs — insurance context
  • Medical record numbers — cross-referenced in AI summaries
  • Health plan numbers — insurance context
  • Account numbers — billing context
  • License numbers — provider license info in referrals
  • Vehicle IDs — accident context in trauma notes
  • Device IDs — implant notes
  • URLs — patient-submitted links to health records
  • IP addresses — remote session logs
  • Biometric IDs — fingerprint or voice print data
  • Photographs — linked media in AI systems
  • Any other unique ID — custom facility identifiers

AI models can create any of these from context. Detection must cover all 18 — not just SSNs and dates.

How to Add Pre-Save Detection

A pre-save PHI check follows five steps:

  1. AI writes the note draft
  2. Note text goes to a detection API before staff sees it
  3. Flagged items are shown in the draft view
  4. Staff reviews the flags during normal note review
  5. Staff saves the note — without flagged items, or with a logged reason

What the system needs:

  • Speed: under 200ms so it does not slow the workflow
  • Coverage: all 18 HIPAA categories plus local patterns such as your MRN format
  • Scoring: items above 85% are auto-flagged; 50–85% need staff review; below 50% are shown for reference only
  • Audit log: log each flagged item, its score, and the reviewer's decision

The audit log gives you direct proof for the HHS risk analysis. It shows you have controls for AI-generated PHI.

Use Case: Pre-Save Detection at a Medical Center

One academic medical center used an AI ambient system for physician notes. A 90-day audit found two mix-up cases. One note had another patient's birth date. A second had a family member's name and SSN from social history.

After adding pre-save PHI detection:

  • All AI drafts were scanned before physician review
  • Average scan time: 47ms — not felt in the workflow
  • Over 90 days: 1,247 items were flagged across 8,400 notes
  • Staff reviewed and resolved 94% of flagged items
  • Zero record mix-up incidents after the launch

The system produces a monthly report. It shows detection rates, review rates, and entity types. This report serves as audit controls proof under HIPAA Security Rule 164.312(b).

Teams building this workflow can use anonym.legal's PHI detection API. It covers all 18 HIPAA categories at sub-200ms latency. See the PHI detection integration guide for setup steps. For end-to-end context, visit the healthcare use cases page.

Sources

Ready to protect your data?

Start anonymizing PII with 285+ entity types across 48 languages.

About this page

We update this page when our platform or the law changes.

Read our founder note for how we work.

Each change shows up in the timestamp at the top.

Related reading

We follow these rules

  • GDPR (EU 2016/679).
  • ISO/IEC 27001:2022.
  • NIS2 (EU 2022/2555).
  • HIPAA safe harbor under 45 CFR § 164.514(b)(2).

Our promise

We do not sell your data.

We do not train models on your text.

We store your files in Germany.

You can delete your account at any time.

You own your work.

Where we run

Our servers live in Falkenstein, Germany.

We use Hetzner. They hold ISO 27001 certification.

All data stays in the EU.

Backups run every day.

Need help?

Email support@anonym.legal.

We reply within one business day.

How we test

We run a full check suite on every release.

Each surface gets its own sweep script and report.

Human reviewers spot-check the output each week.

We track recall and precision on a labelled set.

Bad runs block the deploy.

What we never do

  • We never sell your information to third parties.
  • We never train models on what you upload.
  • We never keep your work after you delete it.
  • We never share keys with any outside firm.
  • We never run ads inside the product.

Plans in plain words

We sell credits, not seats.

One credit covers one short job.

Long jobs use a few credits each.

You can top up at any time.

Unused credits roll over each month.

Read the plans page for current rates.

Who built this

A small team of engineers and lawyers built this.

We ship from Europe and work in the open.

Our founder note spells out why we started.

Where to start

How the parts fit

A browser add-on cleans text inside Chrome.

A Word plug-in handles drafts in Office.

A small desktop tool works on whole folders.

An agent protocol link feeds large models safely.

All four share one core engine and one rule set.

Words from our team

We started this work after a lunch about cookies.

One friend kept getting odd ads on her phone.

We asked why a court file leaked through a draft.

We sketched the first build on a napkin that week.

By month three we had a tiny demo for a friend.

She used it on her first case the next day.

Common questions we hear

Can the tool read scanned PDFs? Yes, with OCR.

Does it work on long files? Yes, in small chunks.

Can I roll my own rule set? Yes, save it as a preset.

Does it run offline? The desktop build runs offline.

Do you keep my files? No, the cloud build wipes after each run.

Will it learn from my work? No, we never train on inputs.

A short tour of the workflow

Upload a file or paste a snippet of prose.

Pick the entities you want gone from the draft.

Choose a method: replace, mask, hash, encrypt, or redact.

Press run and watch the side panel show each hit.

Skim the result and tweak any rule that misfired.

Save the cleaned file or send it to a teammate.