DataSecOps, Explained Through One Passenger Manifest

DataSecOps, Explained Through One Passenger Manifest

Abstract definitions bounce off people. So instead of defining DataSecOps — the discipline, named in August 2021 by Satori Cyber’s Ben Herzberg, of protecting data that keeps changing under users who keep changing — we will run one concrete object through its entire loop: the passenger manifest for a single private charter. Picture a mid-size operator flying a pharma delegation from Mumbai to Geneva next Thursday. The manifest holds eleven rows and nine columns: full legal names, passport numbers and expiry dates, seat assignments, weight figures, catering notes that quietly mention a dietary restriction, a payment reference, an interpreter flag, a dangerous-goods declaration, and the booker’s corporate account. Every stage below acts on that one table. Watch how much machinery a single flight justifies.

Discovery: where does the manifest actually live?

The obvious copy sits in the scheduling database. Then it gets complicated. A PDF went to the handling agent in Geneva. A spreadsheet version sits on the dispatcher’s laptop. The catering company received a trimmed copy by email. The trip’s WhatsApp thread holds photographs of two passports. An analyst once exported the week’s flights to a BI tool nobody has opened since. Discovery is the unglamorous sweep that lists every one of these copies, and on a real trip it usually finds more than this — because protection applied only to the database is protection of one-ninth of the truth. The stage matters more than any tool: you cannot gate what you have not found.

Classification, the DataSecOps way: which columns carry risk?

Now label what was found. On this manifest the tiers fall out cleanly:

  • Direct identifiers: full legal names, passport numbers and expiry dates — the highest tier.
  • Health-adjacent: the catering note about a dietary restriction, which deserves sensitive treatment even though nobody filed it as medical.
  • Financial: the payment reference, tied to a corporate account.
  • Logistical, low-risk on its own: seat assignments, catering counts, weight figures.

Automated classifiers will propose these labels, and operators should let them — Google’s classifier and the open-source Presidio engine both do credible work — but neither is completely accurate, so a person signs off on every tag marked identifier or sensitive before rules depend on it. The interesting judgement call here is the interpreter flag: harmless alone, identifying in combination. Classification is where such decisions get written down instead of staying in someone’s head, and the written record is what survives staff turnover.

Enforcement: rules applied at the instant of the query

This is the stage that separates modern practice from folder passwords. Policies attach to the row, the column, or even the single query, and they evaluate when data is read. The caterer’s account queries the manifest and sees names and seat numbers, but the catering-notes column arrives masked to “standard” unless the passenger opted in. The handling agent in Geneva sees passport data, because border handling requires it. The broker who sold the trip sees booking totals but not one passenger name. The night dispatcher sees everything; the day runner sees rows for today’s flights only, because the rule filters on the departure date. Databases now ship this machinery natively — row-level policies in PostgreSQL since version 9.5 (January 2016), column masking in Apache Ranger since its 0.6 days, tag-based controls on the cloud warehouses — and the engine-specific catch is that each implements it differently, with its own documented gaps.

Egress: controlling what leaves the building

Reading is one thing; exporting is another. DataSecOps treats the exit as its own control point. Bulk exports take a named route, land in a log, and carry a reason. When the 2024 extortion wave tracked as UNC5537 drained customers of the Snowflake platform, the thieves did not hack anything exotic — they authenticated with stolen credentials and ran plain COPY INTO export statements, in a campaign where 79.7% of hit accounts had already been exposed to an infostealer and carried no second factor. A manifest operation with egress logging would have flagged an export of eleven rows to an unfamiliar destination within minutes. The control is cheap; the silence it replaces is expensive.

History: what actually happened, and when

After the flight, someone eventually asks a hard question: who viewed or moved this manifest in the last ninety days? The answer comes from access history, and it arrives with limitations an analyst should know cold, because audit comfort built on wrong assumptions collapses in a dispute. Snowflake’s history view, ACCESS_HISTORY, spans 365 days yet runs as much as three hours behind, and queries that failed never appear in it — so a burst of failed attempts is invisible there and needs another control. pgAudit, the PostgreSQL audit trail, cannot reliably account for what superusers do. OpenLineage fills the pipeline-shaped holes, tracking data as it hops between systems. History is the stage where operators discover their tools’ honest edges, usually for the first time.

Review: where DataSecOps keeps the other stages honest

Two quarters later, the delegation’s trip is long closed, but in the un-reviewed world its access grants live on: the broker’s contract ended, the caterer rotated accounts, a dispatcher changed roles. Review is the standing appointment that retires such leftovers. Access in a mature operation is granted with an expiry from the start — a request names the dataset, its purpose and its lifetime, the owner approves, the grant dies on schedule. One published worked review counted 310 grants of 1,900 (16.3%) dormant beyond ninety days, plus 212 accounts of 940 (22.6%) with more reach than their jobs required. Ratios like that are precisely why the review must revoke, not merely re-affirm. On the manifest’s scale: after the trip closes, nearly everyone’s access should lapse within a week.

Why the loop beats the tool for DataSecOps

Run the manifest through all six stages and a pattern appears: each stage feeds the next. Discovery without classification produces lists nobody can act on; classification without enforcement is stationery; enforcement without egress control guards the front door while the back door stands open; history without review documents negligence in writing. This is why experienced operators distrust any purchase that promises to collapse the loop into one box — overlays such as Satori (now part of Commvault, since the acquisition announced 28 August 2025), Immuta and Privacera are useful, but they accelerate the stages rather than deleting them. The full atlas of the discipline, stage by stage, is the one linked in this sentence: the DataSecOps reference at xOps School.

There is also a commercial angle worth naming. Manifest discipline shows up in customer trust, and trust shows up in the numbers discussed in what actually drives private jet charter pricing — clients paying a premium expect the premium to include discretion. The operational habits behind that discretion pair naturally with how a charter booking actually comes together, because every stage of the trip has a data shadow. Even the market’s quieter corners, such as how empty-leg flights are sold, depend on knowing exactly who may see a passenger’s identity and when. One manifest, six stages, and a discipline that turns a privacy promise from a paragraph in a contract into rows in a log.