CDEP key terms

A branch of applied ethics that examines how data is collected, processed, shared, and used, and evaluates those practices against moral principles such as fairness, respect for persons, and avoidance of harm.

The study of how people ought to act, offering frameworks such as consequentialism, deontology, and virtue ethics that can be applied to justify or critique data practices.

An ethical framework judging an action by its outcomes, so a data practice is right when its overall benefits outweigh its harms across affected parties.

An ethical framework holding that some duties and rules bind regardless of consequences, so certain data uses are wrong even if they produce net benefit.

An approach focused on the character and dispositions of the person acting, asking what an honest, prudent, and fair data professional would do in a situation.

A principle requiring that individuals be treated as autonomous agents whose choices matter, and that those with diminished autonomy receive protection.

The obligation to act for the benefit of others, in data work meaning designing systems and analyses that produce genuine value while limiting foreseeable harm.

The obligation to avoid causing harm, which in data practice means anticipating and preventing foreseeable harms such as exposure, discrimination, or misuse before deployment.

A principle requiring that benefits and burdens of data systems be distributed fairly, with particular attention to whether disadvantaged groups bear disproportionate costs.

Concern with how goods and harms are shared across a population, applied to data when asking who gains from a model and who absorbs its errors.

Fairness of the processes used to make decisions, including transparency, voice for affected people, and the ability to challenge outcomes.

The idea that the legitimate limit on a person's or organization's action is the harm it causes others, often used to justify restrictions on data use.

Agreement to data processing given voluntarily by a competent person who has received and understood clear information about purpose, risks, and alternatives.

A person's right to revoke previously granted permission, after which further processing must stop and stored data must be handled as the applicable rules require.

The rule that data collected for a stated purpose may not be repurposed for incompatible uses without a new lawful basis or fresh consent.

Collecting and retaining only the data actually needed for a defined purpose, reducing exposure, storage cost, and the potential for misuse.

Using data for a purpose different from the one it was originally gathered for, such as research reuse of clinical records, which raises consent and compatibility questions.

Voluntary contribution of personal data for research or public benefit, typically under explicit terms that describe governance and permitted uses.

The set of entitlements individuals hold over their personal data, commonly including access, correction, deletion, portability, and objection to processing.

A person's ability to have personal data deleted when it is no longer necessary, consent is withdrawn, or processing was unlawful, subject to legal exceptions.

The right to receive one's personal data in a structured, commonly used, machine-readable format and to transmit it to another provider.

Any information relating to an identified or identifiable individual, including identifiers and combinations that can single a person out.

Sensitive personal information such as health, biometric, genetic, racial or ethnic origin, political opinion, religion, or sexual orientation, which attracts stricter handling rules.

Irreversibly removing or altering identifying features so that no individual can reasonably be re-identified, taking account of reasonably available means.

Replacing direct identifiers with artificial keys so data cannot be attributed to a person without additional separately held information, while remaining personal data.

The process of linking supposedly de-identified records back to specific individuals, often by combining datasets or using auxiliary information.

A privacy model requiring that each combination of quasi-identifiers in released data be shared by at least k records, so no individual stands out.

A formal guarantee that adding or removing one person's record changes analysis outputs only slightly, typically through calibrated noise, bounding what can be inferred about any individual.

A quantified allowance for how much information about individuals can be released under differential privacy, consumed by each query or release.

Unintended exposure of data to parties who should not have it, whether through breaches, misconfiguration, poor access controls, or careless sharing.

Technical and administrative mechanisms that determine which users or systems may read, modify, or transfer specific data, based on roles, attributes, or need to know.

The security principle of granting each user or process only the minimum access necessary to perform its function, limiting damage from error or compromise.

Assigning data to sensitivity tiers such as public, internal, confidential, or restricted, which then determine handling, storage, and access requirements.

The system of decision rights, policies, and accountability structures that determines who may make which data decisions and how those decisions are enforced.

The operational responsibility for data quality, definitions, and compliant use, carried out by named roles accountable to the governing body.

The obligation to be answerable for data decisions and their effects, including demonstrating compliance and remedying harm when things go wrong.

An independent legal or organizational structure that holds data on behalf of beneficiaries and makes decisions about access according to a defined charter.

Systematic and unfair skew in model outputs that disadvantages particular groups, arising from data, design choices, or deployment context.

Bias that enters a dataset because the world it records already contains unequal treatment, so faithful measurement still reproduces past injustice.

Bias caused by some populations being under-sampled or excluded during data collection, so a model performs poorly or unfairly for those groups.

An attribute used in place of a directly relevant one, such as postcode standing in for ethnicity, which can import discrimination even when protected traits are excluded.

The study of formal criteria, such as equal error rates or calibration across groups, for judging whether a model's outputs treat groups equitably.

A structured independent examination of a model or data system to test performance, fairness, and compliance, and to document findings for accountability.

The expectation that people affected by automated decisions can obtain meaningful information about the logic, inputs, and consequences involved.

The degree to which a person can understand why a system produced a given output, including which inputs mattered and how they combined.

A design method that identifies relevant human values early and treats them as requirements throughout the technical build rather than as afterthoughts.

An approach embedding privacy protections into system architecture and defaults from the outset, rather than adding them after deployment.

Published standards from a professional body that state expected behavior for practitioners, often covering integrity, competence, and duties to the public.

A committee that reviews proposed data projects against ethical and legal criteria before work begins, with authority to require changes or refuse approval.

A prepared sequence of steps for detecting, containing, investigating, and reporting a data breach, including notification to affected people and regulators.