Get Free Quote

Five DataSecOps Beliefs Our Eye Hospital Retired

Health IT

I keep the servers at a mid-size eye hospital, which means I keep the records of people who, for a few weeks, could not read a consent form without help. That detail changes how you think about data. Last quarter our operations committee held a “beliefs audit” — my phrase, borrowed from how clinicians retire outdated practices — and five comfortable sentences about our data got systematically retired. The discipline behind the exercise is DataSecOps — a young label — August 2021 is when Ben Herzberg, running research at Satori Cyber at the time, handed the name to security work that lives inside datasets which never stop changing and user lists which never stop churning — and no standards body has ratified it; the settled name in practice remains data security governance. What follows is what we unlearned, belief by belief, in the order the committee retired them.

Belief one: “the warehouse is encrypted, so the records are covered”

This one died fastest, killed by a single question from a surgeon: “encrypted against whom?” Encryption at rest protects disks and backups from theft — nothing more. Anyone with working database credentials reads everything in the clear, and credentials are exactly what the 2024 Snowflake-customer extortion wave (tracked as UNC5537) abused: in most hit accounts — 79.7% of them — the credentials had already been lifted by infostealer malware, multi-factor authentication was nowhere, and some passwords had sat unchanged for four straight years. The thieves never cracked anything; they logged in. IBM’s annual study reached USD 4.99 million per breach in 2026, the worst figure yet recorded, and none of those incidents were defeated by disk encryption. The committee’s replacement belief: encrypt everything, and then govern every query.

Belief two: “we gave researchers hashed patient IDs — that is anonymous”

We had done this ourselves for a glaucoma-prevalence study and felt rather clever. The committee retired the belief in ten minutes. A hash is deterministic: the same identifier always produces the same string, so hashed medical-record numbers can be matched against a hashed list of guesses — re-identifying patients without ever “breaking” anything. What we had produced was pseudonymisation, the weaker cousin. Law has caught up with the maths. In the September 2025 judgment delivered under docket C-413/23 P, the EDPS against SRB matter, pseudonymised records stop counting as personal data only for a recipient truly unable to reverse the pseudonyms — and for whoever holds the key, the records stay personal all along. Ours is the hospital that holds the key. The replacement belief: call hashed identifiers pseudonymous, and handle them under the same rules as the originals.

Belief three: “the platform we bought enforces everything”

Two years ago we licensed an overlay product — one of the tools that apply rules from outside the database, alongside Immuta and Privacera (the market has since consolidated — Commvault moved to acquire Satori in August 2025, on a preliminary USD 28.3 million basis). The licence felt like protection. It was not. Any path that bypasses the overlay bypasses its rules: a direct connection a resident set up, a table clone a researcher made, a share handed to an equipment vendor. The industry even has a documented anti-pattern for buying such tools before classifying your data — the platform arrives and politely does nothing, because nothing is labelled. Our replacement belief: classify first, then layer the overlay on top of the engine’s own row and column controls, and write every rule as code so console-clicked policies — which evaporate during migrations and copies — stop accumulating.

Belief four: “masked data and de-identified data are the same thing”

The committee’s own ophthalmic photographers contributed the example that retired this one. Masking conceals a value where it sits; de-identification is a decision to release, taken once the leftover re-identification risk has actually been measured — NIST’s SP 800-188 draws the line precisely there. An old academic result still does the teaching: three ordinary facts — postcode, sex and day of birth — made 87% of Americans identifiable on their own. Our cataract-outcomes extract removed names and kept the rest until we ran the numbers ourselves. HIPAA’s Safe Harbor route strips eighteen categories of identifiers, and its small-cell arithmetic teaches the same caution: ZIP codes collapse to 000 whenever the surrounding population runs 20,000 or fewer, and every patient aged 89 or above lands in a single pooled bracket — a ninety-year-old in a quiet county gives themselves away through those two details. Europe’s data-protection board has drafted 2025 guidance to match the spirit: pseudonymisation on its own will normally not be a sufficient measure. The replacement belief: masked data stays inside; only measured, de-identified datasets leave.

Belief five: “our quarterly review meetings are enough”

We had held review meetings for years. The minutes showed attendance; nothing else. When the committee finally asked for evidence — which accounts had been revoked in the last year — the answer was none, and the dormant-access count explained why it mattered: published worked examples run to 16.3% of grants sitting unused for ninety days or longer. An eye hospital accrues the same residue — the visiting researcher whose fellowship ended, the equipment vendor’s support login, the extern from a completed rotation. The replacement belief has teeth now: every grant carries an expiry from birth, requests state their dataset, purpose and duration before an owner approves them, and each review must revoke at least one thing or explain in writing why not. Protection that nobody ever removes is just a growing pile of doors left without locks.

Five DataSecOps replacement beliefs, in one list

The audit ended the way a good ward round ends: with a written plan anyone can read. The committee pinned the five replacement sentences to one page, so that the next administrator, locum analyst or vendor manager absorbs them before the old habits have a chance to reform:

  • Encryption is the floor, not the wall. Disks and backups are encrypted, and every query is still governed, every identity still checked, every credential still rotated.
  • Hashed means pseudonymous. Our keyed identifiers travel under exactly the handling rules of the records they point to, never under an “anonymous” stamp.
  • Tools sit on top of classification. Nothing gets licensed before the sensitive tables are found and labelled, and every rule lives in code, reviewed like code.
  • Masked data stays inside. Only measured, de-identified extracts leave the building, each release carrying a signature and a stated purpose.
  • Reviews revoke, or explain in writing. Every quarter removes at least one grant; a quarter with no removals is treated as a finding, not a success.

What the DataSecOps committee kept, and what an eye hospital should do next

Five beliefs retired, five replaced, and the hospital unchanged in every way patients would notice — except that our evidence now exists. The maturity ladder we use to track it runs ad hoc, classified, enforced, evidenced, adaptive, with the top rung meaning that each new sensitive table turns up pre-gated, ahead of its very first grant, without exception; half of tier-one coverage gets you out of level two, and passing ninety percent coverage while every console-only rule is gone takes you out of level three. The reference the committee keeps returning to is the DataSecOps page at xOps School’s Ops Atlas, which maps the whole loop — find, label, gate, watch, review — without selling anything.

Peer hospitals are, quietly, doing the same audit. Telehealth services for eye care — the subject of our guide to telehealth in ophthalmology — now carry visual-field results and retina images across networks that deserve exactly this scrutiny. Choosing where to have surgery, as our eye-hospital selection guide explains, increasingly includes how a hospital treats your record, not just your cornea. And for the acute side — the walk-in retinal detachment our emergency eye surgery overview describes — speed and privacy have to coexist, which is what the loop delivers. Retire the comfortable sentences before someone else retires them for you; that is the whole discipline, in committee language.