Every certificate you hold rests on a foundation of data: dimensional reports, material certificates, destructive test results, torque traces, furnace logs. When an auditor or investigating authority arrives, they inspect your records before they inspect your parts. If those records can be altered without trace, deleted without detection, or written under a shared login, your certificate is standing on wet paper.

Across two decades of audits in automotive and aerospace, I have watched quality departments discover mid-audit that a test result had been "corrected" by a well-meaning engineer overwriting the original file. The measurement was probably fine. The evidence was not. Information security for quality data is not an IT topic to delegate; it is the discipline of making your evidence believable, and it fails in specific, predictable ways.

The failure modes are few and repeatable: shared accounts, overwritten originals, unverified supplier certificates, untested backups and improvised incident response. Each has a concrete control. This article sets out those controls in the order an auditor will test them.

Access Control on Quality Records

Access control sounds bureaucratic until you trace how quality records actually move through a plant. The CMM programme sits on a network drive with permissions granted to "everyone" because it was simpler a decade ago. The operator at the hardness tester logs in with a generic account. The laboratory technician holds administrator rights on the test-stand PC because the software vendor demanded them at installation. Each arrangement quietly destroys the chain of accountability your certificate depends on.

The principle is simple: named individuals, least privilege, and segregation of duties. Every device that writes data — CMMs, tensile testers, spectral analysers, torque tools — should write under a named account bound to a badge or unique password. Generic accounts such as "lab" or "operator1" must be eliminated from anything producing quality evidence, because when two people share a login, neither is accountable for the record. The person who performs a test should not be the person who can approve or alter the result, and neither should be able to delete raw data.

Practical checks are worth doing quarterly. Pull the user list for your quality data systems and verify each account maps to a current employee; leavers are the classic gap, holding access months after their badge stopped working. Test whether a production operator can reach the calibration records folder — if yes, the permissions are wrong. Challenge vendor demands for local administrator rights on test-stand PCs; a read-only service account is often sufficient. Check whether anyone can edit archive files directly rather than through the application that enforces versioning.

Every measurement written here becomes evidence the moment a certificate depends on it — and evidence without attribution is not evidence.
Every measurement written here becomes evidence the moment a certificate depends on it — and evidence without attribution is not evidence.

Integrity of Test Data: Originals Must Stay Original

Data integrity fails in a handful of recurring patterns, none of which involve hackers. The most common is the overwritten original: an engineer opens the raw CMM export in a spreadsheet to build a customer report, fixes an "obviously wrong" point, and saves over the source file. The second is the retrospective edit: a failed torque trace is re-measured and the failing run replaced in the archive because "the second one is the real result". The third is clock manipulation — the test-stand PC is set back so a late test appears to have run inside the production window.

Defence starts with architecture. Raw data files should be write-once wherever the system allows: written to a location the instrument account can create in but not modify, then hashed or at minimum timestamped by a mechanism outside the tester itself. Reports must be built from copies, never from the source. Where software writes to a database, insist on an audit trail recording who changed what and when, and verify the trail cannot be switched off by the same account that edits results. I have seen systems where the audit log was configurable by the supervisor — which is to say, no audit trail at all.

Detection matters as much as prevention, because a determined insider cannot be fully prevented. Reconcile test counts against production counts: if the line ran 4,200 pieces and the tester logged 3,900 results, evidence is missing, and a missing record damages a certificate as much as a falsified one. Sample the audit trail monthly for edits clustered around shift end, results changed from fail to pass, and deleted rows. Compare file modification timestamps against instrument execution logs. Any mismatch deserves a conversation, not an accusation — but you only find the guilty explanations by asking about the innocent ones first.

Reconstructing an evidence chain

  1. 01Identify the recordPick one submitted result and trace it back to the raw measurement.
  2. 02Verify the instrumentConfirm the device was in calibration at the execution timestamp.
  3. 03Confirm the authorThe named account must map to a person who was present.
  4. 04Test the audit trailNo edits, deletions or timestamp anomalies on the record.
  5. 05Attempt an editTry to modify the raw file under your own account. Success means failure.
A quarterly dry run on one product family: if any step cannot be answered without gaps, the control does not exist.

Supplier Data Exchange

Your certificate covers your parts, but your evidence includes your suppliers'. Material certs, PPAP-style dimensional results, plating thickness reports, calibration certificates from subcontracted labs — all arrive from outside your perimeter, usually by email, and are then pasted into your quality system as if verified. A forged or altered supplier certificate is among the hardest failure modes to detect, because the document looks exactly like a genuine one.

Start by controlling the channel. Certificates arriving as PDFs to a shared inbox, forwarded by a purchaser, carry no chain of custody whatsoever. A supplier portal where suppliers upload under their own authenticated credentials, or a managed file exchange with per-supplier accounts, at least gives you attribution. For critical characteristics — material chemistry, weld qualifications, destructive test results — require the raw laboratory data alongside the summary certificate, and spot-check that the summary matches the raw report. Clerical mismatches are common and usually innocent; consistent, structural mismatches are not.

Verification techniques are cheaper than most plants assume. Ring the issuing laboratory on a published number — not the number printed on the certificate — and confirm the report reference exists; forged certificates frequently invent numbers or reuse genuine ones with altered heats. Watch for inconsistent fonts within a line, altered digits in heat or batch fields, and page numbering that skips. Cross-check certificate batch numbers against the physical batch in your stores. And when a supplier sends suspiciously perfect results — every dimension dead-centre, every chemistry exactly nominal — treat it as a flag, because real processes vary and real reports show it.

What to Measure and What to Check

Security of quality data is auditable in concrete terms, and a small set of standing metrics beats a bulky programme. Track the number of generic or shared accounts still writing to quality data systems, with a target of zero. Track audit-trail exceptions per month — edits to results, deleted records, timestamp anomalies — reviewed by someone other than the person who caused them. Track supplier certificate rejections and their reasons: formatting problems indicate sloppiness, content mismatches indicate risk. Track time-to-revoke for leavers; anything beyond their last working day is a finding waiting to happen.

Periodic deep checks belong in the internal audit calendar. Pick one product family each quarter and reconstruct its evidence chain from raw measurement to customer submission. Pick one supplier certificate at random and verify it against its source. Attempt, as an auditor would, to open a raw data file and modify it under your own account — if you succeed, the control does not exist. These dry runs are uncomfortable, which is precisely why they are valuable.

Backup and recovery belong in this measurement set too, because integrity includes survival. A quality record that exists only on one test-stand PC is one hard-drive failure away from not existing, and "the data is lost" is not a defence anyone accepts during a product investigation. Check that backups run; then check the harder thing: restore one, and confirm the restored data is complete and readable. Untested backups are a comfortable lie.

A missing record damages a certificate as much as a falsified one — absence of evidence is itself the finding.

Incident Handling: When Evidence Is Already Compromised

Despite good controls, something will eventually go wrong: a corrupted file, a suspected falsification, a ransomware event that encrypts the LIMS. What separates a recoverable situation from a certificate-threatening one is whether you can establish, after the fact, which records are trustworthy. That requires knowing your baseline: where raw data lives, what the audit trails cover, and which records you can independently reproduce from retained samples or re-measurement.

Falsification suspicion deserves a defined path, not an improvised one. Preserve the originals immediately, copy nothing onto them, and pull the system logs before anyone with editing rights is told there is a problem. Re-measure retained samples where they exist; physical evidence arbitrates between conflicting records. Decide product disposition on the basis of re-verified data, not the disputed record — if the record is in doubt, its pass result is in doubt too. Document the whole sequence; an investigation into data integrity that itself lacks integrity is a farce.

Ransomware deserves specific mention because it targets exactly the systems quality depends on, and test-stand PCs running older operating systems are notorious soft spots. Segment the quality network from the office network so an infection on a workstation cannot reach the LIMS or the archive. Keep at least one backup copy offline or immutable. Rehearse, on paper at least, how you would answer a customer in week two of an outage: which products were released on data you can no longer produce, and what physical evidence stands behind them. Plants that answer quickly survive; plants that cannot, lose certificates.

Two cultures, two outcomes

Evidence gets hidden

  • Overwritten originals stay concealed
  • Failing runs replaced in the archive
  • Mismatches explained away locally
  • Falsification surfaces during the audit

Evidence gets defended

  • Self-reported mistakes thanked publicly
  • Raw files untouchable by procedure
  • Mismatches escalated at the tester
  • Problems surface on a Tuesday, in time to fix
The difference between a self-reported overwrite and a concealed edit is the difference between a correction and an audit finding.

Building It Into Daily Practice

None of this needs a grand programme. It needs the quality function to treat data records with the care it treats the physical part, and to say so in the procedures it already writes. Access rules belong in the record-control procedure, not buried in an IT policy nobody in the lab has read. Integrity rules — no edits to raw files, reports built from copies, discrepancies escalated — belong in work instructions at the tester, where the operator actually looks. Supplier verification rules belong in purchasing and incoming inspection documents.

Training should cover the failure modes, not the policy paragraphs. Show people the overwritten original, the backdated timestamp, the heat number that does not match the label on the bin. Practitioners who have seen how easily evidence becomes unbelievable will defend it without being told. And when someone self-reports a mistake — the engineer who saved over a raw file and confesses — thank them publicly. The alternative culture, where mistakes get hidden, is the culture that produces the falsification you discover during an audit instead of a Tuesday.

The certificate on your wall makes a promise that your data must be able to prove. Access control, integrity protection and disciplined supplier exchange are how you keep that proof standing when somebody finally, inevitably, comes to check it.