Book Appointment

When Everyone Sees Everything: A DataSecOps Diagnosis

Clinic Operations

Every clinic that has grown quickly knows this feeling: nobody can quite say who can see what anymore. The scheduling sheet opens for the whole building. The reporting warehouse grants a role called something like ALL_ACCESS to people whose jobs never needed it. Saying it out loud sounds like an accusation, so nobody says it out loud. This piece is the conversation, framed the way clinicians already think — DataSecOps treats the situation as a diagnosis with numbers, then a treatment, then a re-check, and the label itself is just the vendor world’s shorthand for security work that moves with changing data (Ben Herzberg of Satori Cyber coined it in August 2021; standards bodies never did, and the older name, data security governance, still describes the same work).

Three complaints that show up before any diagnosis

Over-granting announces itself through symptoms that feel unrelated until you line them up.

  • Reports break whenever someone’s role changes, because permissions were wired to individuals rather than purposes.
  • Auditors ask “who approved this access?” and the room goes quiet, because the honest answer is “no one remembers.”
  • Everyone maintains a private spreadsheet copy of anything important, precisely because the official access feels either impossible or frightening.

Sports clinics see the same pattern in overuse injuries: nothing dramatic happened on any single day, yet tissue damage accumulated quietly through repetition. Permissions behave identically — every project, merger, and departed colleague leaves a little behind, and the total compounds while nobody watches.

Taking the pulse: the DataSecOps numbers behind the feeling

Feelings cannot be treated; figures can. A worked example from a real warehouse estate shows what measurement looks like, and the four numbers are worth memorising because they fit on one slide. Coverage led the sheet: of 1,240 sensitive columns, 1,085 carried protection — an 87.5 percent share, encouraging until you notice the unguarded remainder is exactly where trouble hides. Exposure came second: 17.1 percent of all reads touched sensitive values in plain text, meaning roughly one query in six met unmasked data. Over-exposure came third: among 940 principals, 212 could touch material beyond their duties — a 22.6 percent slice of the workforce. And dormancy came fourth: 310 of 1,900 grants had gone unused for ninety days, 16.3 percent of the estate, every one of them a door nobody had thought about in months. Much like how doctors evaluate knee pain by ordering measurements before naming a cause, systematic evaluation of knee pain starts with evidence, and so does this: name the percentages, then treat. The framework behind numbers like these is written up in the DataSecOps guide at Ops Atlas, and it repays a careful read before the first audit arrives.

How a clinic arrives here without anyone deciding to

The causes are boring, which is why they survive. Roles get copied rather than designed, so the analyst template quietly carries billing rights. A single global pipeline role gets created “temporarily” for an integration and becomes load-bearing forever — the classic anti-pattern where one account reads everything because untangling it felt harder than leaving it. Rules accumulate in a vendor console, clicked by people who have since left, unversioned and unreconstructable, which is why console-only policies now fail audits outright. Nobody sets an end date on anything, because expiry requires an owner and owners require meetings. And the quarterly review, if it exists at all, revokes nothing, which converts it from a control into a ritual. Add the ordinary churn of healthcare staffing — rotations, locums, a departing manager whose twelve grants outlive their twelve projects — and the estate fills with permissions that exist only because deletion was never anyone’s job. None of this requires malice. It requires only time plus absent ownership, both of which every growing clinic has in abundance.

The DataSecOps treatment: doors that close themselves

The plan works in strict order, and skipping steps is how projects stall. Classification comes first, because a rule can only be as good as the label underneath it — and scanner output still needs a human pass, since no classifier catches everything or avoids over-flagging. Policy comes second: row filters and column masks attached to those labels, so the front desk sees appointments without medical history, physios see clinical fields without financial ones. Third, every new grant carries a purpose and an expiry, and unusual requests pass through a ticket with a risk sign-off before a deadline-stamped permission issues. Fourth, the review calendar fires quarterly with revocation expected rather than exceptional — dormant grants from the pulse-check get cut in the same sitting. Engines help but quibble, so read the documentation before promising dates: Snowflake’s row policies famously do not block inserts, meaning the write path needs separate attention, and a few platforms treat each new table as a fresh governance problem. Automation carries the load throughout, because doors that need a human to close them stay open precisely when everyone is busy.

Six weeks later: watching the numbers move

Re-run the same four measurements after the first month and the curve tells you whether treatment is working. Coverage of sensitive columns should climb toward the maturity ladder’s second rung, which opens above fifty percent of tier-one tables under real policy; the third rung wants more than ninety percent and, tellingly, zero console-only rules left anywhere. Plain-text read share should fall as masks take effect. Over-exposure shrinks fastest, because expiry plus revocation attacks it directly. Dormancy is the satisfying one — those 16.3 percent of grants evaporate the first time the review actually revokes. A clinic that cannot show these numbers moving is doing theatre; one that can has converted a complaint into a managed metric, and managed metrics stay managed. Expect the first re-measurement to feel slow and the second to feel unfair — protection work loads before it lifts, the same way strengthening exercises ache before they help.

Keeping the weight off after treatment

Recovery matters less than maintenance, in knees as in access. The habits are small and repetitive: an owner named for every dataset, requests phrased with purpose and duration, a review that expects cuts, and policies living in version control where their history can be read rather than guessed. Clinics that normalise these find audits become boring, in the best possible way — the evidence already exists, dated and filed, and the annual visit shrinks from an examination into a formality. Patients benefit even when they never notice; the reason trusted knee-care information matters to patients applies equally to their records, because trust in a clinic is built from exactly these invisible disciplines. And when someone asks whether all this really matters for everyday care — the sort of everyday concern behind why knees hurt when activity changes — the honest answer is that records follow patients for decades, long after the current appointment book is archived.

Over-granting never announces itself as an emergency, which is what lets it grow for years. The cure is unglamorous: measure, label, close, and keep closing. Six weeks of that routine turns “everyone sees everything” into a slide you would be happy to show an assessor — and, honestly, into a workplace where the quiet question of who is watching finally has a precise answer.