Skip to content

Reference

Privacy and AI governance glossary

The vocabulary of global privacy work, defined in the way practitioners actually use it. Every term links to a shareable anchor so you can point colleagues at a definition.

Showing 135 of 135 terms.

A

Access control(Least privilege)Data security
Granting each account only the access necessary for its role, and reviewing that access periodically. Over-permissioned collaboration platforms are the most common source of accidental internal exposure.
AccountabilityPrivacy law
The obligation not only to comply but to be able to demonstrate compliance through documentation, assessments, contracts, training records and evidence that controls operate.
AccuracyPrivacy law
Keeping personal data correct for its intended use, and correcting or erasing inaccurate data when required.
Adequacy decisionPrivacy law
A determination by a competent authority that another jurisdiction provides an essentially equivalent level of data protection, allowing transfers to proceed without additional safeguards.
AI deployerAI governance
Under the EU AI Act, a party using an AI system under its authority, except personal non-professional use; role changes can create additional duties.
AI governanceAI governance
The set of policies, roles, assessments and controls that determine how AI systems are selected, built, deployed, monitored and retired, and who is accountable for their outcomes.AI privacy and data readiness
AI impact assessmentAI governance
A structured assessment of an AI use case covering purpose, data, affected individuals, accuracy, bias, transparency, oversight, security and fallback, carried out before deployment.
AI literacyAI governance
Knowledge and understanding enabling informed use of AI and awareness of its opportunities, risks and possible harms, tailored to roles and context.
AI providerAI governance
Under the EU AI Act, a party developing or having an AI system or general-purpose AI model developed and placing it on the market or putting it into service under its name, subject to the Act’s definition.
AI risk registerAI governance
A maintained record of AI risks, affected people, controls, residual risk, decisions, owners and review triggers.
AI tool register(AI inventory)AI governance
A maintained list of AI systems in use, with owner, purpose, data categories, vendor, risk rating and review date. It is the AI equivalent of a processing record and the first thing an auditor asks for.
Algorithmic biasAI governance
Systematic disparity in AI outputs across groups, arising from training data, labelling, feature selection, deployment context or feedback loops. Relevant to both discrimination law and fairness under privacy law.
AnonymisationData security
Irreversibly altering data so an individual is no longer identifiable by any reasonably likely means. Genuinely anonymised data falls outside data protection law, which is why the threshold is high.
Attribute-based access control(ABAC)Data security
An access model evaluating attributes of the subject, resource, action and environment against policy.
Audit trailData security
A chronological record of relevant actions supporting investigation and accountability; logs themselves need access and retention controls.
AuthenticationData security
Verifying the identity of a user, service or device before access; authentication alone does not determine which actions are permitted.
AuthorisationData security
Determining the actions and resources an authenticated or anonymous actor may access, based on a policy.
Automated decision-makingPrivacy law
A decision taken about an individual by automated means without meaningful human involvement. Where it produces legal or similarly significant effects, most regimes require safeguards, transparency and often a route to human review.

B

Backup retentionData security
How long recoverable copies persist and how deletion or hold requirements are applied when data is restored.
Binding Corporate Rules(BCRs)Privacy law
Internal data protection rules adopted by a corporate group and approved by a supervisory authority, allowing intra-group international transfers without separate contracts.

C

ContractorPrivacy law
A California privacy-law role with contractual and certification requirements restricting the use and disclosure of personal information received from a business.
Controller(Data Fiduciary / Business)Privacy law
The organisation that determines the purposes and means of processing personal data. The controller owes the primary duties to individuals and regulators, even where a supplier performs the processing.
Cross-context behavioural advertisingPrivacy law
Targeting advertising to an individual based on their activity across distinct sites, applications or services that are not owned by the business with which they intentionally interact.
Cryptographic erasureData security
Rendering encrypted data unrecoverable by securely destroying all relevant keys; assurance depends on encryption design, remaining key copies and plaintext copies.

D

Dark patternPrivacy operations
Interface design that nudges people toward choices against their interests, such as unequal accept and reject options in a consent banner. Regulators increasingly treat these as invalidating consent.
Data classificationData security
Assigning sensitivity levels to information so that handling, sharing and retention controls can be applied consistently and automatically.
Data intermediaryPrivacy law
Under Singapore’s PDPA, an organisation processing personal data on behalf of another under a written contract; direct obligations and role-specific exceptions matter.
Data inventory(Data map)Privacy operations
A view of what personal data exists, where it is stored, which systems process it and who owns it. A RoPA describes activities; an inventory describes locations and assets.
Data lineagePrivacy operations
A record of data origins, transformations and destinations that helps trace disclosures, derived records and dependencies.
Data loss prevention(DLP)Data security
Controls that detect and prevent sensitive data leaving approved locations or channels, based on content inspection, classification labels or context.
Data minimisationPrivacy law
The requirement that personal data be adequate, relevant and limited to what is necessary for the purpose. In practice it is a design constraint on forms, logs, telemetry and data warehouses.
Data portabilityPrivacy law
The right to receive personal data provided by the individual in a structured, commonly used, machine-readable format and, where technically feasible, have it transmitted to another organisation.
Data protection impact assessment(DPIA / PIA)Privacy operations
A structured assessment of a processing activity's risks to individuals and the measures that reduce them, carried out before high-risk processing begins.DPIA triage tool
Data protection officer(DPO)Privacy operations
A designated individual responsible for advising on, and monitoring, compliance with data protection law. Several regimes mandate appointment based on sector, scale or risk, and require independence and reporting to senior management.
Data retention schedulePrivacy operations
A documented set of retention periods by record type, with the legal or business justification and the technical enforcement point where deletion occurs.
Data sharing agreementPrivacy operations
An agreement between controllers setting out the purposes, scope, security and responsibilities for personal data shared between them, including how rights requests and incidents are handled.
Data subject(Data Principal / Consumer)Privacy law
The individual to whom personal data relates and who holds the rights granted by the applicable law.
Data subject access request(DSAR / SAR)Privacy operations
A request from an individual for a copy of their personal data and information about how it is processed. Handling it well depends on identity verification, search scope, redaction and a defensible response clock.
De-identificationData security
Removing or obscuring direct identifiers. It sits between pseudonymisation and anonymisation in strength and is defined differently across regimes, so the label alone does not determine legal treatment.
Deletion verificationPrivacy operations
Evidence that an approved deletion action reached the intended systems and copies, including exceptions and backup expiry rather than an unsupported promise.
DerogationPrivacy law
A narrowly framed exception to a general rule, such as particular GDPR transfer exceptions; it is not a default replacement for regular safeguards.
Differential privacyData security
A mathematical framework limiting how much an analysis reveals about any one participant; protection depends on parameters, composition and implementation.

E

Encryption at rest and in transitData security
Protecting stored data and network traffic using cryptography. It reduces breach impact and, in several regimes, affects whether individuals must be notified after an incident.
ExfiltrationData security
Unauthorised transfer of information out of an approved environment, whether by an attacker, compromised account or misuse.
ExplainabilityAI governance
The ability to describe, in terms a person affected can understand, the logic and main factors behind an automated decision or recommendation.

F

FairnessPrivacy law
Processing personal data in ways that are not unjustifiably detrimental, misleading or unexpected for the people affected.
Fine-tuningAI governance
Further training of an existing model on additional data; it can introduce new privacy, security, bias and licensing risks.
Fundamental rights impact assessment(FRIA)AI governance
An assessment addressing impacts on fundamental rights; specific EU AI Act duties apply only to defined deployers and systems when relevant provisions apply.

G

General-purpose AI model(GPAI)AI governance
A model displaying significant generality and able to competently perform a wide range of distinct tasks, subject to the EU AI Act’s definitions and exceptions.

H

HallucinationAI governance
An AI output that is unsupported or incorrect but presented plausibly; human verification is especially important for consequential decisions.
Harvest now, decrypt laterData security
Collecting encrypted information today with the expectation of decrypting it when future capabilities improve; data longevity affects migration priority.
HashingData security
Mapping input to a fixed-length digest. Predictable values can be guessed, so hashing alone does not establish anonymity.
High-risk AI systemAI governance
A legal classification under the EU AI Act based on specified uses and conditions, not simply a high score in an internal risk assessment; staged applicability must be checked.
Human contestabilityAI governance
A practical route for a person to challenge an AI-influenced outcome and obtain meaningful review by someone able to change it.
Human oversight(Human in the loop)AI governance
Meaningful review by a person with authority and information to change an AI-informed outcome. Rubber-stamping does not meet the standard where a regime requires human involvement.

I

Inherent riskPrivacy operations
Risk considered before the proposed controls are credited, to support a consistent assessment and a clear mitigation rationale.

J

Joint controllersPrivacy law
Parties that jointly determine processing purposes and means. Their arrangement allocates duties but does not extinguish individuals’ rights against either party.

K

Key managementData security
The creation, protection, rotation, use, recovery and destruction of cryptographic keys with defined ownership and access.

L

Lawful basisPrivacy law
The legal ground relied on for a processing activity. Under the GDPR the six bases are consent, contract, legal obligation, vital interests, public task and legitimate interests; other regimes use narrower or differently framed lists.
Legitimate interests assessment(LIA)Privacy operations
A documented three-part test — purpose, necessity and balancing — used to justify processing under the legitimate interests basis and to evidence that individual rights and expectations were considered.
LinkabilityData security
The ability to associate records or actions with the same person or entity, even without knowing their name.

M

Machine unlearningAI governance
Techniques intended to remove the influence of specified training data; effectiveness and resulting model behaviour require validation.
Membership inferenceAI governance
An attack attempting to determine whether a record or individual was represented in model training data.
Model cardAI governance
Structured documentation of a model’s intended use, limitations, evaluation, provenance and operational constraints; it is not itself a compliance certificate.
Model driftAI governance
A change in data or relationships over time that reduces a model’s relevance or performance and can alter risks to people.
Model inversionAI governance
An attack attempting to infer sensitive attributes or reconstruct aspects of training data from a model’s behaviour.
Model output disclosureAI governance
The risk that an AI system reveals personal or confidential information in a response, whether from retrieved context, memory, training data or an over-broad system prompt.
Model training dataAI governance
The data used to train or fine-tune a model. Privacy questions include lawful basis, minimisation, retention, whether personal data can be extracted from the model, and whether deletion requests can be honoured.
Multi-factor authentication(MFA)Data security
Authentication using factors from different categories, such as possession and knowledge; phishing-resistant methods offer different assurance from one-time codes.

N

Necessity and proportionalityPrivacy operations
Evaluating whether processing is needed for a stated purpose and whether its extent and impacts are justified relative to less intrusive alternatives.
Notifiable breachPrivacy operations
A personal data breach meeting the applicable threshold for reporting to a regulator or to affected individuals. Thresholds and clocks vary by jurisdiction, so a single global rule rarely works.

O

Onward transferPrivacy law
A subsequent transfer by the first recipient to another recipient or location, which must remain consistent with applicable safeguards.
Opt-out preference signal(Global Privacy Control)Privacy law
A browser or device signal communicating an individual's choice to opt out of sale or sharing of personal information. Several US state laws require covered businesses to honour it.

P

Personal data(Personal information)Privacy law
Any information relating to an identified or identifiable individual. Identifiability includes indirect routes: an account number, device identifier or combination of attributes can be personal data even when no name is present.
Personal data breachPrivacy operations
A security incident leading to accidental or unlawful destruction, loss, alteration, unauthorised disclosure of, or access to personal data. Availability incidents count, not just disclosures.
Post-market monitoringAI governance
Ongoing collection and review of evidence about an AI system’s performance and risks after deployment, with corrective-action processes.
Post-quantum cryptography(PQC)Data security
Cryptographic algorithms designed to resist attacks by both classical and quantum computers; migration requires asset and dependency planning.
Privacy by designPrivacy operations
Building privacy safeguards into systems, processes and default settings from the outset rather than retrofitting them after a design is fixed.
Privacy control evidencePrivacy operations
Records demonstrating that a control operates, such as access-review signoffs, deletion logs or tested response workflows, not merely a policy statement.
Privacy engineeringPrivacy operations
Translating privacy goals into technical requirements, architecture, implementation tests and evidence of operating controls.
Privacy incident registerPrivacy operations
A restricted log of incidents, facts, risk decisions, notifications, actions and follow-up, including documented reasons not to notify.
Privacy maturity modelPrivacy operations
A scale describing how developed a privacy programme is, typically from ad hoc to optimised, used to set realistic improvement targets rather than pass or fail judgements.Privacy readiness assessment
Privacy notice(Privacy policy)Privacy operations
The external-facing explanation of how an organisation processes personal data. It must reflect real practice; a notice describing processing that does not happen is itself a transparency failure.
Privacy threat modellingPrivacy operations
Examining how a system could cause privacy harms through data flows, inference, linkability, exposure or misuse, before selecting controls.
Processing registerPrivacy operations
An activity-led record linking owners, purposes, data categories, recipients and safeguards; local legal record requirements may add fields.
Processor(Data intermediary / Service provider)Privacy law
An organisation that processes personal data on behalf of, and under the documented instructions of, a controller. Processors owe direct duties on security, sub-processing and assistance under most modern regimes.
ProfilingPrivacy law
Automated processing of personal data to evaluate personal aspects of an individual, such as performance, economic situation, health, preferences, reliability, behaviour, location or movements.
Prompt injectionAI governance
An attack where instructions embedded in content processed by a model cause it to ignore its intended constraints, potentially disclosing data it can access or taking unintended actions.
PseudonymisationData security
Processing personal data so it can no longer be attributed to an individual without additional information kept separately and protected. Pseudonymised data remains personal data.
Purpose creepPrivacy operations
The gradual expansion of processing beyond the purposes originally communicated, usually through incremental internal reuse of an existing dataset.
Purpose limitationPrivacy law
The requirement to collect personal data for specified, explicit and legitimate purposes and not to process it further in a manner incompatible with those purposes.

R

RansomwareData security
Malware or attacks that extort an organisation through encryption, disruption or threatened disclosure; privacy assessment must consider availability and exposure.
Re-identificationData security
Linking purportedly de-identified information back to an individual through auxiliary data, inference or distinctive patterns.
RecipientPrivacy law
A person or organisation to whom personal data is disclosed; specific legal definitions can exclude certain authorities receiving data for a particular inquiry.
Records of processing activities(RoPA)Privacy operations
A structured register of processing activities recording purposes, categories of data and individuals, recipients, transfers, retention and security measures. It is the backbone artefact most other privacy work depends on.RoPA starter kit
Red teamingAI governance
Structured adversarial testing of a system’s failure modes and safeguards; findings must lead to owned remediation and retesting.
Residual riskPrivacy operations
Risk remaining after controls are applied. Acceptance requires an authorised owner and does not waive legal duties.
Restriction of processingPrivacy law
Limiting further use of stored personal data while a qualifying dispute or condition is addressed; it is distinct from deletion.
Retrieval-augmented generation(RAG)AI governance
An architecture in which a model retrieves content from a document store at query time. Privacy risk comes from the permissions on that store: retrieval inherits whatever over-sharing already exists.
Right to erasure(Right to be forgotten)Privacy law
The right to have personal data deleted where a specified ground applies, subject to exemptions such as legal obligations, defence of legal claims and freedom of expression.
Right to objectPrivacy law
A right to challenge particular processing. Under GDPR, direct-marketing objections have different rules from objections to legitimate-interest processing.
Risk appetitePrivacy operations
The level and types of risk an organisation is prepared to accept within legal constraints and its governance framework.
Role-based access control(RBAC)Data security
Granting permissions through defined roles rather than individual ad hoc grants, with role assignment and review controls.

S

Sale of personal informationPrivacy law
A statutory concept covering specified disclosures for monetary or other valuable consideration; definitions and exceptions depend on the applicable law.
SaltingData security
Adding a unique value before hashing to reduce reusable lookup attacks; the required design differs between passwords and data-matching use cases.
Schrems IIPrivacy law
The CJEU’s 2020 judgment invalidating EU–US Privacy Shield and confirming that use of SCCs requires assessment of effective protection and additional safeguards where necessary.
Secure development lifecycle(SDLC)Data security
A development process integrating security and privacy requirements, review, testing, release controls and ongoing maintenance.
Secure disposalData security
Destroying or sanitising information and media using methods appropriate to sensitivity, technology and evidence requirements.
Security incident responseData security
The documented process for detecting, triaging, containing, eradicating and recovering from an incident, including the decision points that trigger privacy breach assessment.
Sensitive data discoveryData security
Scanning structured and unstructured repositories to locate sensitive data types. Value depends on detection tuning and a remediation decision model, not on scan volume.
Service providerPrivacy law
A California privacy-law role for a party processing personal information on a business’s behalf under a qualifying written contract and use restrictions.
Shadow ITData security
Technology used by teams without organisational review or approval. It matters for privacy because unreviewed tools receive personal data outside inventories, contracts and retention controls.
Sharing of personal informationPrivacy law
Under California law, specified disclosure for cross-context behavioural advertising, whether or not money changes hands; statutory exceptions apply.
Significant Data FiduciaryPrivacy law
A designation under India’s DPDP Act that can trigger additional duties when the relevant provisions apply; designation and commencement must be checked.
Singling outData security
Isolating a person’s record or behaviour within a dataset, which can create identifiability without a direct name.
Special category data(Sensitive personal data)Privacy law
Categories of personal data that attract additional conditions before processing, typically including health, biometric and genetic data, racial or ethnic origin, political opinions, religious beliefs, trade union membership and data about sex life or sexual orientation.
Standard Contractual Clauses(SCCs)Privacy law
Pre-approved contractual terms adopted by a regulator or authority that provide a safeguard for personal data transferred to a country without an adequacy decision.
Storage limitation(Retention limitation)Privacy law
The requirement to keep personal data in identifiable form no longer than necessary. A retention schedule only satisfies it when deletion actually happens at a technical enforcement point.
Sub-processorPrivacy operations
A third party engaged by a processor to carry out part of the processing. Most regimes require authorisation, flow-down of terms and notice of changes.
Supplementary measuresPrivacy operations
Technical, contractual or organisational safeguards added to a transfer mechanism where necessary to address destination-law and practice risks.
Synthetic dataData security
Artificially generated data intended to reproduce useful properties of a source dataset; leakage and identifiability still require testing.
System cardAI governance
Documentation of an AI system’s components, use context, evaluation and safeguards, extending beyond the underlying model.

T

Third partyPrivacy law
A party outside the relationship specified by the applicable law. The meaning differs between privacy regimes and must be assessed in context.
Third-country transferPrivacy law
A disclosure or availability of personal data to a recipient outside the relevant legal territory where transfer rules apply; remote access can be relevant.
TokenisationData security
Replacing a sensitive value with a substitute token, often backed by a separate lookup service; reversibility means it is not automatically anonymisation.
Training data provenanceAI governance
Evidence of where training data came from, how it was collected and licensed, and what privacy restrictions or transformations apply.
Transfer impact assessment(TIA / TRA)Privacy operations
An assessment of whether a transfer mechanism such as Standard Contractual Clauses delivers protection that is essentially equivalent in the destination country, considering local law, practice and supplementary measures.
TransparencyPrivacy law
Making processing understandable to the people affected, including purposes, recipients, rights and meaningful information about relevant automated decisions.

V

Vendor risk assessment(Third-party due diligence)Privacy operations
A structured review of a supplier's data handling, security posture, sub-processing, location and contractual commitments before and during engagement.Vendor privacy quick check
Vulnerability managementData security
Identifying, prioritising, remediating and verifying weaknesses in systems based on exposure and impact.

W

Z

Zero trustData security
A security model that assumes no implicit trust based on network location and continuously verifies identity, device posture and authorisation for each access request.