Healthcare
Your staff summarise long notes, rewrite discharge instructions in plainer language, and draft prior-authorisation appeals. Itzal finds the PHI in that text on the clinician's own device and stops it before it is sent — without PHI ever reaching us.
What’s specific here
Clinical language is not like other corporate text. A tool tuned for finance will either flood your privacy office with noise or miss the disclosure that actually matters.
A phone number in a signature block is not PHI. The same number beside “discharge summary” and a date of birth is. Itzal fires HIPAA severity only when an identifier co-occurs with health context terms — and that term list is yours to edit, because a hospital directory would otherwise generate constant noise.
Safe Harbor requires removing all eighteen identifier categories, including dates more specific than year and geography below state level. Staff who believe they de-identified a note usually have not.
Cloud-inspection DLP makes your security tool a recipient of the very data you are protecting — a second BAA and a second place to be breached. Itzal’s servers cannot hold PHI, because they never receive it.
Anything that adds perceptible delay to documentation gets escalated out of the environment within a week, and an uninstalled agent protects nothing. Itzal hard-releases the submit rather than making anyone wait.
De-identification
It is still PHI. Three of the eighteen Safe Harbor identifiers are present, and every one of them would survive a find-and-replace on the patient’s name.
62-year-old male, admitted 3 March, MRN 4471102, resident of Bellevue, presenting with…
A date more specific than year, a medical record number, and a geographic subdivision below state level. Invented data.
The eighteen Safe Harbor identifiers
Deployment
Enforcement before tuning is how a DLP deployment gets reversed. Each stage below exists to earn the right to the next one.
An enterprise tenant under a BAA, and good enough that people actually prefer it. Prohibition without an alternative reliably produces unmonitored use on personal phones.
One department, two to four weeks. Nothing is blocked. The point is to measure false positives against real clinical language rather than against a test corpus.
Health context terms, plus exception lists for test data, training material and the sample records that live in every training environment.
Watch the dismissal rate. High click-through means precision needs work — it does not mean the warning needs to be louder.
HIPAA findings are blocked to anything without a BAA, while the sanctioned tool stays the path of least resistance.
On speed. Itzal budgets a p99 of 15 ms on the pattern path, and is designed to hard-release the submit at 250 ms whatever the detector is doing. Clinical staff have the least tolerance in any organisation for a tool that slows them down, and the rate of those released events is a health metric in your console rather than something hidden.
Before you size other controls
These gaps matter disproportionately in healthcare, because a screenshot of a patient list and an attached census export are both ordinary workflows rather than edge cases.
Not currently inspected
Straight answers
Itzal’s servers never receive PHI, because detection and redaction happen on the device. Many healthcare customers still execute a BAA as a matter of policy, and that is straightforward — but the architecture is designed so that a breach of Itzal’s infrastructure could not expose PHI.
No. HIPAA severity fires only when an identifier co-occurs with health context terms, and that term list is editable per organisation. Without it, a hospital’s own directory would generate constant false positives.
Yes — that is the intended outcome. Sanction a tool under a BAA, then use policy to make the sanctioned destination the path of least resistance while blocking PHI to everything else.
Not legal advice. Whether a particular disclosure is reportable is a determination for your privacy officer and counsel, not for a vendor.
The useful demo is the one run against notes that read like yours. Bring a de-identified sample and the AI tools your staff already use.