Cognitive Liberty Glossary | Cognitive Liberties

Cognitive liberty: The principle that a person should retain meaningful freedom over thought, attention, memory, identity, mental privacy, lawful inquiry, and cognitive self-direction.
Forum internum: The internal domain of thought, conscience, and belief, distinguished from outward expression or conduct.
Mental privacy: The ability to keep thoughts, mental states, neural data, and sensitive cognitive inferences from unwanted observation, profiling, storage, sharing, or use.
Intellectual privacy: The protected space needed to read, search, draft, research, and form ideas away from unwanted observation or interference.
Search privacy: Protection for query content, query histories, associated identity, and inferences derived from information-seeking behavior.
Cognitive autonomy: The practical ability to direct one’s own thinking and decisions rather than being covertly steered by another actor or system.
Personal agency: A person’s capacity to make and act on choices, including the practical ability to refuse, change course, or choose alternatives.
Safety poison pill: A policy pattern in which a narrow, emotionally compelling harm is used to obtain a broader surveillance or control mechanism whose reach or secondary uses exceed what the harm requires.
Function creep: The expansion of a system, dataset, or authority from its original purpose into new uses.
Expansion ratchet: The institutional tendency for exceptional powers to broaden or persist because of sunk costs, incentives, weak sunsets, secrecy, and the value of retained data.
Epistemic due process: Notice, reasons, context correction, appeal, and review when an automated system materially changes access to knowledge or classifies an inquiry.
Algorithmic Discoverability Due Process: A proposed framework giving publishers notice, reason categories, evidence, appeal, timely correction, and transparency when automated systems materially suppress visibility.
Hypernudging: Dynamic, highly personalized choice architecture that uses large-scale data and machine learning to adapt influence in real time.
Persuasive technology: Technology intentionally designed to influence attitudes or behavior. Cognitive-liberty concerns increase when influence is hidden, personalized, difficult to escape, or based on intimate profiling.
Harmful capability uplift: The additional practical ability a system gives a user to execute a harmful act compared with ordinary public information and existing tools.
Over-refusal: A safety failure in which an AI system blocks or avoids a benign request because it resembles a harmful one.
Query-local intervention: A safety response limited to the immediate request and not retained as a persistent judgment about the user.
Identity separation: An architecture that prevents the party processing a query from also learning or retaining the user’s real identity when identity is unnecessary.
Oblivious HTTP: A protocol that separates the party seeing the client’s network identity from the party processing encrypted request content.
Private Information Retrieval: A cryptographic method that lets a client retrieve a database record without revealing which record was requested.
Differential privacy: A mathematical approach that adds controlled uncertainty to aggregate results so an individual contribution cannot be reliably inferred.
Anonymous credential: A token or proof that demonstrates an attribute—such as an age threshold—without revealing unnecessary identity information.
Digital identity: A credential, account, identifier, or set of verified attributes used to connect a person with digital access or activity.
Pseudonymity: Participation under a stable name or identifier that is not automatically linked to a legal identity.
Censorship drift: The gradual expansion of a narrow restriction or safety mechanism into broader control over lawful expression, inquiry, or access.
Soft censorship: Invisible omission, reframing, downranking, throttling, or answer shaping that changes what people encounter without an explicit ban or refusal.
Intermediary liability: Rules that hold hosts, platforms, providers, or distributors responsible for user activity, often creating incentives to remove more content than the law requires.
Sovereign internet: A nationally controlled network architecture designed to centralize routing, filtering, name resolution, or cross-border connectivity; resilience claims must be separated from censorship capability.
Transnational repression: Attempts by a state or aligned actor to intimidate, surveil, threaten, censor, or punish people and institutions outside its territory.
Epistemic conformity: The narrowing of questions, sources, and expressed beliefs toward what people think an observer, platform, institution, or majority will accept.
Distributed archive: A collection preserved across multiple institutions, formats, locations, languages, or legal jurisdictions so one failure or order cannot erase it.
Internet fragmentation: The division of the global internet into technically or legally incompatible networks that restrict cross-border reach, interoperability, or access.
AI memory: Persistent information used by an AI system to influence later interactions, including stored facts, preferences, summaries, inferences, or relationship context.
Provenance: Information about where a record came from, what changed it, which source is authoritative, and how a derivative relates to the original.
Neurotechnology: Technology that measures, interacts with, or influences nervous-system activity, including brain-computer interfaces and neural sensing.
Algorithmic suppression: A reduction in visibility, reach, recommendation, retrieval, or citation caused by automated ranking or classification without a formal ban on the underlying material.
Financial censorship: The use of banking, payment, advertising, insurance, fundraising, or other economic chokepoints to restrict a person or institution’s ability to publish, organize, or reach an audience.
Language censorship: Control exercised through forbidden terms, compulsory euphemisms, controlled vocabulary, semantic narrowing, or penalties attached to the words used to describe events and ideas.
Translation censorship: The omission, alteration, selective availability, or systematic mistranslation of information as it moves between languages, preventing audiences from receiving equivalent evidence.
Reader privacy: Protection for what a person reads, searches, views, borrows, annotates, or asks about, including the associated identity, timing, and inferences.
Relational privacy: Protection for the social graph revealed by communication metadata: who interacts, when, how often, and within which communities.
Censorship-resistant archive: An archive designed so no single order, failure, custodian, vendor, format, jurisdiction, or technical chokepoint can silently erase the only surviving record.
Personalized censorship: Individualized filtering, ranking, refusal, or framing based on identity, location, history, language, inferred traits, or predicted vulnerability.
Training-data censorship: Knowledge loss or bias introduced when material is excluded, mislabeled, underrepresented, or made inaccessible during the construction of AI training corpora.
Knowledge resilience: The capacity to create, discuss, fund, translate, distribute, discover, read, teach, verify, preserve, and restore information despite political, commercial, technical, or military pressure.