PHI Data Loss Prevention: Policies That Reduce Risk Without Blocking Care
DLP that blocks a clinician emailing a referral will be switched off within a month. Starting in monitor mode, tuning to real clinical traffic, and blocking only what is unambiguous.
GuardsArm Team
Security Experts
DLP has a poor reputation in healthcare for a specific reason: it is usually deployed in blocking mode against rules written for a corporate environment, it immediately stops a clinician sending a legitimate referral, and it is disabled or exempted into irrelevance within weeks.
The technology is not the problem. Sequencing is.
Why generic rules fail here
A rule that flags "a document containing a name, a date of birth and a diagnosis" describes an attack in a bank and a Tuesday in a hospital. Clinicians legitimately send PHI constantly — referrals, discharge summaries, specialist consultations, transfers.
The distinction DLP must make is not "does this contain PHI" but "is this PHI going somewhere it should not, in a volume or manner that does not fit the work". That is a harder question and it requires knowing what normal looks like first.
The sequence that works
What is worth blocking
Keep this list short and defensible. Everything on it should be something no clinician would ever legitimately do.
| Rule | Why it is safe to block |
|---|---|
| Bulk export above a threshold to personal cloud storage | No clinical workflow requires this |
| PHI to a known personal email domain in volume | Distinguishable from a single patient-requested copy |
| Unencrypted PHI to an external recipient where encryption is available | The encrypted path exists and works |
| Copying a patient database to removable media | Should route through an approved export process |
| PHI posted to a public site or paste service | Never legitimate |
Everything else — a single referral to a named external clinician, an attachment to a payer, a specialist consultation — should alert and log, not block.
Cover the channels that matter
Email is where DLP is usually deployed and is rarely where the largest volume leaves.
Generative AI assistants deserve specific mention. Staff paste clinical text into them to summarise or rewrite, which is a disclosure to a third party and frequently to a model provider's retention. Decide the policy deliberately, provide a sanctioned tool if you can, and cover the unsanctioned ones in DLP.
Measuring it properly
- Alerts per 1,000 messages — should fall sharply through tuning
- False positive rate, sampled by hand, because the tool cannot self-assess
- Blocked events with a successful override — high numbers mean the rule is wrong
- Coverage by channel, not just by rule count
- Time to review a genuine alert, which is the whole point of generating it
Where to start
Turn discovery on and find out where PHI actually sits. Most organisations are surprised — by the departmental file share, the spreadsheet of patients on a shared drive, the mailbox with a decade of attachments. Reducing standing PHI is often cheaper and more effective than policing its movement.
GuardsArm deploys and tunes DLP in clinical environments, including the monitor-mode baseline that makes enforcement survivable. Book a scoping call.
Written by GuardsArm Team
Our team of cybersecurity experts brings decades of combined experience in penetration testing, compliance auditing, and incident response. We're dedicated to helping organizations strengthen their security posture.
Take the next step on this topic
Talk to the GuardsArm team about how these services apply to your environment.
