AI Glossary / Data governance and quality / Intermediate

Data integrity

Definition

Data integrity is the condition in which data remains consistent with its intended content and applicable rules, and is protected against improper alteration or destruction. In cybersecurity, the emphasis is on preventing or detecting unauthorized modification. In databases, integrity also includes constraints such as valid values, unique identifiers, and valid references. These related uses should be distinguished from the broader question of whether every recorded fact is true or fit for a particular purpose.

What it means

Integrity means being able to rely on the data's preservation and defined relationships. A file should not change unnoticed, and an order should not refer to a nonexistent customer where the schema prohibits that. Different checks support different aspects of integrity. A cryptographic digest can compare bytes with a trusted reference, while a foreign key checks a relationship. Neither alone proves that the customer is real, the order was authorized, or the information is current.

How it works

Define the properties that must be preserved, then apply suitable controls across collection, storage, processing, and transfer. Database constraints enforce declared rules; access controls limit who can change data; transactions coordinate related updates; integrity checks and logs help detect changes. Backups and recovery procedures support restoration after damage. The controls must work together, and their scope should be documented. A valid local update can still create a business inconsistency if the wider process or cross-system relationship is not covered.

Three practical examples

• Database relationships: an order's customer reference is checked against the customer table according to the declared foreign-key rule.
• File preservation: a stored artifact is compared with a trusted digest to detect unexpected content changes.
• AI reproducibility: dataset and model versions are recorded and verified so an evaluation can be tied to the artifacts actually used.

Think of it this way

Think of maintaining a ledger whose entries cannot be changed casually and whose totals and references follow clear rules. The analogy is limited because digital integrity spans bytes, identities, transactions, and distributed copies, and no single ledger check establishes every aspect of factual accuracy.

Worked example

A toy database records Order O7 for Customer C2 and an amount of 50 units. A foreign key confirms that C2 exists, and a CHECK rule confirms that the amount is nonnegative. Both pass. If the correct customer was actually C3, the record can still violate factual accuracy while satisfying these integrity constraints. Separately, a later unauthorized change from 50 to 500 is an integrity incident even though 500 remains a valid nonnegative number. Logs and change controls address a risk the range check cannot detect. This original example shows why integrity controls must be selected for specific threats and rules rather than treated as one universal correctness test.

Illustrative case study

Fictional example: Pine Analytics publishes a versioned evaluation dataset and records its manifest. A later analysis produces unexpected results, so the team compares the actual input artifact with the manifest and reviews the transformation version. It also checks that the labels were validly assigned, which requires evidence beyond file equality. This invented case illustrates the complementary roles of artifact integrity and semantic quality, without claiming that a matching hash establishes trustworthy labels.

Common misconceptions and mistakes

Do not equate integrity with confidentiality. Data can be publicly readable yet protected against unauthorized changes, or secret yet corrupted. Do not assume encryption alone provides all needed integrity guarantees; the mechanism and configuration matter. Also, a backup is not automatically a verified good copy, and a passing constraint does not prove factual truth. “Integrity” in a database-design discussion may focus on different properties from the same word in a security assessment, so state the intended scope.

Business use

Integrity supports dependable operations, reporting, and evidence. AI systems amplify the need because altered training data, mismatched model files, or inconsistent feature records can affect results without obvious failures. Identify critical data properties and assign controls that can actually test or preserve them. Maintain traceable versions and correction processes, and verify recovery. Review transformations and integrations as well as storage: data can remain intact on disk while being incorrectly mapped or changed by an authorized but faulty process.

People also ask

Is data integrity the same as data accuracy?

They overlap, but integrity checks can preserve an inaccurate original or enforce rules that do not establish real-world truth.

Does a foreign key protect file bytes?

No. It enforces a defined database relationship, not byte-level preservation.

Can data have integrity without confidentiality?

Yes. Public data can be protected against improper modification while remaining openly readable.

Check your understanding

A model file matches its trusted recorded digest, but it was trained using incorrect labels. What has the digest check established?

Answer: It supports that the file matches the recorded artifact. It does not establish the quality of the training labels or the model's suitability for a task.

Related terms

• Availability
• Data Quality
• Data corruption
• Entity integrity
• Referential integrity
• confidentiality
• constraint
• checksum
• authentication
• transaction
• audit trail
• provenance

Learn next

list the integrity properties of one dataset at three levels: unchanged artifacts, valid structures and relationships, and trustworthy business processes. Map a specific control to each property and write down what that control cannot establish. Then test a correction and a recovery scenario against those expectations.

More in Data governance and quality

• Data entry - Data entry is the process of recording information in a computer system so it can be stored, processed, and used.
• Data exfiltration - Data exfiltration is the unauthorized transfer of information out of a system or controlled environment.
• Data governance - Data governance is the system of decision rights, responsibilities, policies, and oversight through which an organization directs how its data is managed...
• Data lifecycle - The data lifecycle is the sequence of activities through which data is planned, obtained, processed, used, maintained, shared, preserved, and eventually...
• Data minimization principle - The data minimization principle requires personal data to be sufficient and relevant for a defined purpose while limiting it to what that purpose needs.
• Data quality dimensions - Data quality dimensions are distinct aspects used to describe and assess data quality.

Sources

Sources: NIST CSRC Glossary, Data Integrity, https://csrc.nist.gov/glossary/term/data_integrity ; NIST, Data Integrity: Identifying and Protecting Assets Against Ransomware and Other Destructive Events, https://www.nist.gov/publications/data-integrity-identifying-and-protecting-assets-against-ransomware-and-other ; PostgreSQL 18, Constraints, https://www.postgresql.org/docs/18/ddl-constraints.html ; National Digital Stewardship Alliance, Checking Your Digital Content, https://www.digitalpreservation.gov/ndsa/working_groups/documents/NDSA-Fixity-Guidance-Report-final100214.pdf . Primary glossary, guidance, and documentation excerpts consulted. The article explicitly distinguishes security, database, and semantic-quality uses; examples are original.

Previous: Data ingestion Next: Data-intensive computing