GASP AICF

Search controls and profiles

Search by control ID, name, domain or profile

APP-015 Processing Integrity

Tier 2+ProviderDeployerGPAI Model ProviderManaged Service Provider

Description

A documented processing specification states, for each service that transforms customer data, the expected inputs, the transformation rules and the required outputs with their delivery timing. Inputs are validated against that specification for completeness and accuracy before processing. Records that fail validation are rejected or quarantined rather than processed. Processing and delivery failures are detected, logged with enough context to identify the affected records and corrected under a defined process that records what was corrected and when. Stored inputs and outputs are protected against undetected alteration. They are reconciled against each other at defined intervals, with every exception investigated to a recorded outcome.

Rationale

Nothing in the library said that a system produces complete, accurate and timely output, so the entire SOC 2 processing integrity category sat unmapped while most service-provider audit reports include it. APP-011 holds input validation at the API boundary as a security control and the prohibition on sensitive data in error responses; APP-015 holds completeness and accuracy of the processing itself, which is a different test on the same request. APP-012 holds the integrity of the software artefacts that do the processing, not of the data passing through them. AIG-027 asks the same question of a probabilistic system through confidence thresholds and fallbacks. Record counts, control totals and checksums are the usual mechanism; the specification is what the counts are tested against, which is why a control without one cannot fail.

Applicability (9 profiles)

SaaS AI Providerstablerequiredcore
Enterprise AI Deployerstablerequiredcore

The processing specification covers the input pipelines the deployer controls, which is where Art.26.3 input data relevance is tested.

GPAI Model Providerstablerequiredcore
High-Risk Provider (EU)stablerequiredcore
Public Body Deployer (EU)stablerequiredcore

The processing specification covers the input pipelines the deployer controls, which is where Art.26.3 input data relevance is tested.

DORA ICT Provider (EU)stablerequiredcore
HIPAA Business Associate (US)stablerequiredrole duty

164.312(c) makes an integrity standard and a detection mechanism explicit obligations over the data itself. APP-015's reconciliation is the mechanism; its gap under this overlay is improper destruction, which the standard names alongside alteration.

NIS2 Cloud Provider (EU)stablerequiredcore

Framework Mappings (10)

HIPAA-164.312.c.1Integritypartial
HIPAA-164.312.c.2Mechanism to Authenticate Electronic Protected Health Informationfull
HIPAA-164.312.e.2.iIntegrity Controlsinformative
SI-11Error Handlinginformative
SI-15Information Output Filteringfull
PI1.1Preparation of Complete and Accurate Informationpartial
PI1.2System Processing Completeness and Accuracyfull
PI1.3System Processing Outputsfull
PI1.4Error Outputsfull
PI1.5Stores Inputs and Outputsfull

Evidence (3)

policydocumentmanual

Processing specification for a service that transforms customer data, stating the expected inputs and their validation rules, the transformation rules and the required outputs with their delivery timing.

Example: Billing run processing specification v2.4, approved 2026-05-18, defining the input file format, the per-field validation rules, the rating logic and the delivery window for each output.

Test: Request the processing specification for a sampled service. Verify: (1) it states the expected inputs and the validation rule applied to each, (2) it states what the service does to those inputs, (3) it states the outputs, their format and the time by which they are delivered, (4) it names an owner and an approval date, (5) the validation rules in the specification are the rules the running service applies on a sampled record.

logtechnicalautomated

Processing log for the period showing the records rejected or quarantined at validation, the processing and delivery failures and the identifiers of the records each event affected.

Example: Pipeline exception log, August 2026, with 1,412 quarantined records and 6 delivery failures, each carrying the batch identifier and the failing rule.

Test: Obtain the processing log for the period. Verify: (1) records that failed validation were rejected or quarantined rather than processed, (2) each failure event identifies the records affected, (3) delivery failures against the specified timing are recorded, (4) every event in the period reconciles to an entry in the correction record or to a documented reason for no action, (5) a deliberately malformed test record is rejected and appears in the log.

recorddocumentmanual

Reconciliation and correction record showing the periodic comparison of inputs against outputs, the exceptions found and the correction applied to each.

Example: Monthly reconciliation record for the billing pipeline, August 2026, with input and output control totals per batch, four exceptions and the correction and re-run for each.

Test: Request the reconciliation and correction records for the period. Verify: (1) reconciliation ran at the defined interval for every service in scope, (2) it compares a control total or count between input and output rather than confirming the job finished, (3) each exception has an investigation outcome and, where a correction was made, what was corrected and when, (4) no exception is open beyond the defined period without a recorded reason.

Questions (3)

boolean

Is there a documented processing specification for each service that transforms customer data?

The specification is what an integrity check is tested against. Without one, a completeness check can only confirm that the job ran.

multi

Which processing integrity controls are in place?

Inputs validated for completeness and accuracy before processingRecords that fail validation rejected or quarantined rather than processedProcessing and delivery failures detected and logged with the records they affectedA defined correction process that records what was correctedOutputs checked against the specification before deliveryStored inputs and outputs reconciled against each other at defined intervalsNone of the above

Quarantine and correction are the pair that matters: a pipeline that drops bad records silently passes a completeness check on what it kept.

select

How are processing exceptions handled?

Every exception is investigated to a recorded outcome within a defined periodExceptions are investigated and the outcome is recorded, without a defined periodExceptions are reviewed in aggregateExceptions are visible in logs and acted on when noticedExceptions are not tracked

Options run from strongest to weakest. This measures whether an exception reaches a conclusion, not whether it is visible.