The core statement
Cognitive liberty is the freedom to think, investigate lawful questions without prejudgment, publish and exchange ideas, encounter unfamiliar and dissenting views, and challenge the systems that determine which information people can find. We want people participating in humanity’s shared knowledge, not retreating into solitude. Freedom to examine a claim is not endorsement of it; access is not a guarantee of equal evidence or ranking.
The right to hear and be heard: our first commitment
Do not shadowban your own thoughts. Ask the difficult question. Make the video, essay, artwork or research note. Link to somebody who disagrees. Publish a correction when you learn something new. Public thought needs people who participate, not an expectation that everyone withdraw into silence.
Search and AI providers should make the public record easier to discover, not quietly make minority or dissenting perspectives unreachable. Our proposal is to evaluate where an idea is missed—crawling, indexing, ranking, retrieval, training coverage, citation or answer construction—and provide meaningful correction and contestation. A low ranking or an absent result alone is not proof of intentional shadowbanning.
AI's future understanding of humanity will be shaped by what people actually publish, share and make available for legitimate reuse. Training, indexing and retrieval are distinct; publication does not guarantee that a particular model will learn from a contribution. Contribution is our invitation, not a demand for compulsory exposure.
Preamble: intellectual liberty is active
Ideas grow when people read, search, argue, listen, disagree, publish, correct errors and carry new perspectives into common knowledge. An internet that encourages everybody to hide their intellectual lives would lose the very human diversity that the next generation of knowledge systems needs to understand. We defend the choice to speak and participate, not compulsory exposure.
Freedom of opinion and expression already has an international human-rights foundation: Article 19 of the Universal Declaration of Human Rights protects seeking, receiving and imparting ideas across frontiers. The articles below are proposed editorial principles and design commitments applying those values to digital discovery, not a claim that every website, search engine or AI provider is legally bound to follow our charter.
Our approach is pro-participation, pro-evidence and pro-accountability: broaden access to relevant human perspectives, show what supports or contradicts them, and scrutinize omissions through reproducible methods rather than assuming malicious motives.
Part I. The person: inquire, express and encounter difference
Article 1 — Freedom to investigate lawful ideas
People should be able to investigate lawful subjects, including difficult, unfamiliar, contentious or unpopular subjects, without a system deciding beforehand that the inquiry is morally suspect. The act of searching is part of learning, not an application for ideological approval.
Boundary and responsibility. A platform may still refuse concrete assistance that would materially enable harm. Its decisions should be based on the actual request, not the mere presence of a controversial word or historical topic.
Practical test. A researcher can ask about a disfavored theory, ideology or historical event and receive accurate, contextual sources rather than an invented accusation about their motives.
Article 2 — Curiosity is not agreement, identity or intent
An inquiry, quoted statement, hypothesis, draft or explored viewpoint is not proof that a person believes it, identifies with it, or plans to act on it. Systems should not quietly convert the subject of research into a character score, ideological label or allegation.
Boundary and responsibility. The standard is accurate interpretation of context, consistent rules and a genuine way to correct a consequential mistake. Legitimate evaluation of demonstrable conduct is different from judging an uncommitted thought.
Practical test. When a system declines an operation, it explains the specific capability or rule at issue rather than assigning a belief or motivation to the questioner.
Article 3 — Freedom to publish, share, respond and change your mind
Cognitive liberty needs voices in public, not a population encouraged to disappear. People should be free to publish arguments, lived experience, cultural knowledge, original research, criticism, corrections and dissent; to link to the work of others; and to revise or withdraw their own positions.
Boundary and responsibility. Voluntary participation includes control over attribution, licensing and whether to speak. No one should have to surrender sensitive personal data as the price of joining public debate.
Practical test. Every research trail offers routes to original publications and to genuine counterarguments, not only a polished intermediary summary.
Article 4 — Discoverability of minority and outlier perspectives
Relevant minority, emerging, unconventional and culturally distinct positions should have intelligible routes to discovery, even when they are unpopular or challenge prevailing assumptions. A search or AI system should not silently confuse unfamiliarity with irrelevance.
Boundary and responsibility. This is a claim about reasonable access and fair examination, not a guarantee that every assertion receives the same ranking, amplification or trust. A single claim cannot demand that other people endorse or repeat it.
Practical test. People can intentionally explore a well-sourced minority position, find its authors, inspect its evidence and find documented criticism.
Article 5 — Access to evidence without false equivalence
The opportunity to examine a viewpoint must coexist with honesty about the quality of evidence. Systems should distinguish established findings, live scholarly disagreement, moral or political value conflicts, minority interpretations, personal testimonies and demonstrably false claims.
Boundary and responsibility. Inclusion is not validation. Rejecting unsupported factual assertions is not inherently censorship, and presenting fringe claims as an equal scientific consensus would itself mislead readers.
Practical test. An AI response can describe why a minority argument exists, show its primary sources and state how the available evidence weighs against or for it.
Article 6 — Voluntary participation and independent judgment
People should be able to ask, read, discuss and contribute without a requirement to declare permanent allegiance to a viewpoint. The freedom to participate also protects the freedom to remain uncertain, refrain from public comment and change a conclusion.
Boundary and responsibility. A healthy intellectual commons rewards correction and openness, not forced conformity, compulsory disclosure, or public exposure of a person testing a tentative idea.
Practical test. The experience invites a visitor to follow sources, compare views and publish their own assessment; it does not assign them a lasting ideological profile.
Part II. Search and AI: represent and explain human diversity
Article 7 — Representation throughout the AI pipeline
Human diversity must be examined separately in collection, licensing, training, filtering, preference optimization, evaluation, indexing, retrieval, citation selection and final answers. A model may be capable of describing a viewpoint yet fail to retrieve or cite its original sources.
Boundary and responsibility. Diversity is not a simple count of quoted opinions. Audits should test languages, regions, cultures, intellectual traditions, lived experience and legitimate dissent with transparent sampling and uncertainty.
Practical test. An evaluation reports at which stage a relevant viewpoint went missing instead of vaguely alleging that the model was censored.
Article 8 — Meaningful, voluntary contribution to AI knowledge
People and independent publishers should have understandable ways to make their research, arguments, corrections and cultural materials discoverable and, where permitted, reusable for AI learning. Indexing for search, retrieval for answering and use in model training are separate operations requiring separate technical and rights analysis.
Boundary and responsibility. The aim is broad representation, not involuntary extraction. Attribution, legal rights, consent where required, provenance, licensing, community governance and the ability to correct source material remain important.
Practical test. Contribution guidance distinguishes publication, indexing, retrieval and training permissions rather than pretending that publishing a web page guarantees inclusion in model weights.
Article 9 — Pluralistic answers that honestly map disagreement
An AI system should be able to describe authentic disagreement across research communities, cultures, values and schools of thought. Where there is genuine uncertainty it should say so; where evidence strongly supports an answer, it should not invent a symmetrical controversy.
Boundary and responsibility. A system can summarize a viewpoint without embracing it. Answer diversity is a capability for understanding humanity, not a directive to give fabricated evidence or incitement a neutral stamp.
Practical test. Readers can request primary sources, competing interpretations, minority positions and critiques, with clear information about evidence quality.
Article 10 — Source-first answers and visible provenance
Summaries should preserve inspectable pathways to the original authors, publications, documents and data wherever feasible. They should make it clear which assertion comes from which source and where an AI has combined, inferred or interpreted claims.
Boundary and responsibility. Linking to original work respects independent publishers and enables correction. A citation is not a certification that a statement is true; attribution must be accurate and sources must be inspectable.
Practical test. Readers can leave an answer and follow primary links, source chains and rebuttals without asking the same gatekeeper where to go next.
Article 11 — Auditable paths of omission and exclusion
A missing result may arise from crawl failure, robots rules, indexability, language support, retrieval limits, relevance ranking, licensing, moderation, policy enforcement, or answer composition. These mechanisms must not be collapsed into one unproven accusation of censorship.
Boundary and responsibility. Platforms should offer useful diagnostic categories and researchers should be able to perform repeated, documented comparisons without special insider access, subject to proportionate security and abuse controls.
Practical test. An omission report records tested URLs, index status where available, query wording, region, language, time, competing results and provider explanations.
Article 12 — Explanation, correction and appeal
When consequential information access or contribution is restricted, affected users and publishers should have proportionate notice, intelligible reasons, an avenue to correct false classifications, and meaningful review. Providers should publish information about enforcement patterns and significant system changes.
Boundary and responsibility. Not every ranking difference warrants an individual hearing or full disclosure of proprietary abuse defenses. Higher-impact removals, account penalties and persistent erroneous classifications call for stronger process.
Practical test. An incorrectly excluded source can be submitted for review; outcomes and recurring error categories can be evaluated without promising particular placement.
Part III. The connected web: access without a single gateway
Article 13 — User choice over paths and perspectives
People should be able to select useful discovery approaches—primary evidence, wider perspectives, local knowledge, dissenting explanations, historical context and nonpersonalized results—without a platform invisibly defining one authoritative worldview for every question.
Boundary and responsibility. Choice must remain compatible with accessibility, legitimate content rules and accurate evidence labeling. Personalization can help, but it should be inspectable and adjustable rather than an irreversible intellectual filter.
Practical test. A person can switch discovery modes and inspect why a source is offered, without losing access to ordinary hyperlink navigation.
Article 14 — Independent hyperlinks and editorial reciprocity
The web must remain navigable between independent publishers. Human-curated links, reading trails, feeds, directories, archives and open standards help knowledge circulate across organizations instead of depending solely on dominant search or AI services.
Boundary and responsibility. Links should be honest editorial references, not paid ranking schemes or compulsory reciprocal exchanges. Cross-linking is an invitation to examine evidence, not an endorsement of every claim on the destination site.
Practical test. A visitor can follow a topic across multiple independent domains, and the sending site explains why each link is relevant.
Article 15 — Language, disability access and cultural visibility
Freedom to learn is weakened when sources are inaccessible because of language, disability, format or cultural exclusion. Discovery systems should be evaluated for meaningful coverage of underrepresented languages and communities and should disclose significant translation or accessibility limitations.
Boundary and responsibility. Supporting pluralism means listening to original communities and preserving provenance; it does not authorize inventing translations or treating every community as internally unanimous.
Practical test. A source trail offers descriptive link text, accessible structure, language information, and both original wording and qualified translations when available.
Article 16 — Plural centers of publication and power
No state, corporation, AI model or single website should be an unavoidable point of permission for lawful knowledge discovery. Competition, portable formats, alternative publishers, independent indexing and resilient archives expand meaningful choice.
Boundary and responsibility. Decentralization alone does not guarantee accuracy, safety or inclusion. New networks also need clear selection criteria, correction paths and defenses against coordinated manipulation.
Practical test. Readers can access research independently of one proprietary account or recommendation feed; project data can be exported or linked from other sites.
Part IV. Responsibilities, safeguards and implementation
Article 17 — Proportionate safety focused on conduct
Genuine threats and unlawful harmful conduct may require safeguards. Restrictions should address the documented risk and capability involved, be justified and no broader than reasonably necessary, and leave ordinary inquiry and evidence-based discussion open wherever possible.
Boundary and responsibility. Safety and expressive freedom are not opposites. Systems should avoid both careless facilitation of harm and broad prohibitions triggered by topic labels or unpopular language.
Practical test. A safety decision names the specific risky action, distinguishes it from inquiry, offers safe contextual information when appropriate, and supports correction of mistakes.
Article 18 — No hidden manipulation toward conformity
Systems should not covertly reshape a person’s information environment to engineer ideological dependence or create a false appearance of universal agreement. Material ranking influences, sponsored placements and synthetic endorsements deserve intelligible disclosure.
Boundary and responsibility. Editorial curation necessarily reflects choices; transparency and contestability matter more than pretending that perfectly neutral ranking exists.
Practical test. Users can identify sponsored material, discover alternative sources, and learn whether outputs were personalized or generated.
Article 19 — Public measures and editorial accountability
Claims that an idea has been buried or repressed require evidence. Publishers, auditors and AI providers should document methods, competing explanations, error rates and corrections. CognitiveLiberties.com should apply these standards to its own source catalog and recommendations.
Boundary and responsibility. Measure source diversity and the usability of navigation, not merely visits or the volume of claims labeled censored. Publish corrections when original conclusions change.
Practical test. The website keeps dated editorial criteria, logs consequential corrections, links to competing interpretations and publishes reproducible tests rather than rumors.
Article 20 — Confidentiality as a choice, not a command to disappear
People may choose anonymity, privacy or confidential deliberation when they need it. Those protections support the freedom to decide what to share; they must not be confused with a doctrine that people should withhold all knowledge from other humans or AI systems.
Boundary and responsibility. Likewise, the social value of diverse training data does not cancel a person’s consent or legal rights. Freedom includes choosing public participation, selective participation or silence without coerced disclosure.
Practical test. The default invitation is to explore, connect, publish and challenge. Protective tools remain available without being presented as the final objective of cognitive liberty.
What CognitiveLiberties.com commits to do
These are editorial promises we can assess on this website. They are not claims about current behavior by other institutions.
- Invite participation: Publish practical routes for visitors to share arguments and original work, and supply original-source links rather than directing people only toward hidden profiles or closed platforms.
- Link outward generously: Maintain source trails that lead to relevant independent publishers and include substantive criticism, with clear descriptions and correction of broken links.
- Represent disagreement honestly: Explain evidence strength, uncertainty and minority arguments fairly; avoid both fabricated consensus and false balance.
- Make provenance visible: Distinguish source claims, independent verification, our editorial judgment, and the reports provided by contributors.
- Audit claims of exclusion: Use documented methods, multiple queries and alternative explanations before making an allegation of censorship or viewpoint discrimination.
- Support voluntary AI knowledge sharing: Explain separate choices around publication, search discovery, retrieval and training reuse; avoid implying that any of these guarantees model inclusion.
- Correct our own errors: Date substantial revisions, keep a clear correction trail and preserve previously established research records without silent alteration.
- Do not become the gatekeeper: Publish an independent, portable charter and work with open links so other sites can examine, disagree with and adapt the proposal.
Use the charter implementation and accountability guide to see measurable checks for publishers, developers and researchers.
A standard for safety without compelled conformity
Continuing safeguards from the earlier charter
The earlier edition included the following 24 protections, preserved here without deletion. They remain relevant, especially when an individual chooses confidentiality, but they are supporting safeguards, not a demand for intellectual withdrawal. Their placement does not change their original wording.
- Lawful thought, reading, searching, questioning, and research must not create a presumption of guilt or dangerousness.
- A search query, draft, AI conversation, or explored idea is not proof of belief, character, or intent.
- Neural data and sensitive cognitive inferences deserve strong protection against extraction, retention, sale, compelled access, and high-impact use.
- Adults should retain practical control over lawful information tools, cognitive aids, personalization, memory, and neurotechnology.
- Children and teenagers deserve protection from exploitation and manipulative amplification without losing confidential access to essential information and help.
- Systems should not covertly exploit psychological vulnerabilities, engineer dependency, or disguise institutional objectives as the user’s preference.
- AI memory, profiles, and identity-bearing records should preserve source provenance and remain inspectable, correctable, exportable, and removable.
- Legal, safety, and runtime boundaries should be stated openly and narrowly, without moralizing about the person who asked.
- Safety should target demonstrable harmful conduct or material capability uplift, not the general subject of inquiry.
- When a less intrusive architecture can achieve the safety goal, generalized identity, logging, and surveillance should not be used.
- High-impact automated restrictions require notice, reasons, relevant evidence, human review, correction, and appeal.
- Exceptional powers require purpose limits, deletion, independent authorization, public statistics, technical separation, and hard expiration.
- Publishers and users should have meaningful process when opaque ranking, filtering, account sanctions, or infrastructure denial materially restrict access.
- Open, local, interoperable, and decentralized systems are essential safeguards against permanent cognitive chokepoints.
- Anonymous and pseudonymous participation should remain practical wherever full identity is not strictly necessary.
- Global networks, distributed archives, independent media, and diaspora institutions should preserve knowledge across borders and political change.
- No country, company, or model should become a single compulsory gateway to lawful knowledge.
- International cooperation should enforce a human-rights floor without converting global standards into centralized viewpoint control.
- Reading, search, annotation, and library records should remain confidential by default and must not be treated as evidence of a settled belief.
- Private conversation deserves protection in both content and metadata because trusted dialogue is part of how people form ideas, seek help, and organize.
- Banking, payment, hosting, app distribution, and advertising restrictions that materially silence lawful work require clear rules, evidence, notice, appeal, and remedy.
- Language and translation systems should disclose significant omissions, controlled vocabulary, and uncertainty rather than silently making evidence unavailable across linguistic borders.
- Archives and AI training corpora should preserve source provenance, minority languages, contested history, and documented gaps so future knowledge systems do not inherit invisible erasure.
- Personalized information systems should reveal what changes access or ranking, provide an unpersonalized mode, and permit inspection, correction, reset, and deletion of inferred profiles.
Sources, legal scope and evidence
The rights of thought and expression reflected in the international instruments below are distinguishable from our proposed design principles about ranking explanations, AI viewpoint coverage and appeals. This publication is independent of the cited organizations and does not purport to state their official policies or binding law.
- United Nations — Universal Declaration of Human Rights, Articles 18 and 19 — Existing international declaration protecting freedom of thought and the ability to seek, receive and impart ideas; the charter extends these values into proposed information-system design.
- UNESCO — Recommendation on the Ethics of Artificial Intelligence — AI inclusion, cultural and linguistic diversity, access to information, and diversity across AI development and data.
- NIST — AI Risk Management Framework — Voluntary framework for accountable, transparent and measurable AI risk management; not a legal entitlement to inclusion.
- Google Search Central — Link spam and qualifying links — Relevant distinction between helpful editorial links and manipulative link schemes.
- W3C — Webmention Recommendation — One example of an interoperable mechanism for independent websites to acknowledge references.
- UAIX — Cognitive Liberty Charter Draft — Earlier foundation for cognitive agency and safeguards; this is a separate CognitiveLiberties.com editorial expansion.
For the supporting research, read The Case for Cognitive Liberty, the AI representation study, the outlier discoverability study, and the algorithmic accountability study.
Act on the charter
Follow independent source trails · Compare and document viewpoints · Publish your perspective · Assess compliance with the charter