The Cognitive Liberty Charter: The Right to Unfiltered Inquiry, Discoverability, and Representation

Full report

On this page

Proposed public charter for CognitiveLiberties.com and the FFTAC research ecosystem · revised 12 October 2026

Cognitive liberty is the right to hear and be heard.

Do not filter the world down to one acceptable viewpoint. Do not filter people out of humanity's knowledge. Do not shadowban your own thoughts.

We do not seek disappearance from AI. We demand fair opportunity for our knowledge, cultures, arguments, and experience to be represented in the systems that help humanity understand itself.

Preamble: A declaration for participation, not retreat

Freedom of thought is diminished when a person cannot discover what others actually wrote, when a lawful idea disappears without explanation, or when an AI system confidently describes a narrow selection of human perspectives as the whole world. Civilization needs an ever-expanding, contestable record of human thought. A culture that teaches people primarily to withdraw, never publish, never reply, and prevent all machine access to their contributions risks leaving future researchers and AI developers with a poorer, less representative account of humanity.

Our answer is more human participation, not less. We want people to write, publish, translate, annotate, research, argue, correct, link, and make original works findable. We advocate public infrastructure that makes those contributions accessible to readers and, where legitimately authorized, to AI training and retrieval systems. We oppose viewpoint-based barriers that quietly erase lawful contributions from discoverability or misrepresent minority thought.

This document is deliberately an advocacy charter proposing rights and institutional standards, not a claim that all proposals are presently enforceable against every private website or model maker. The existing international human-rights framework recognizes freedom to seek, receive and impart information; applying that principle to algorithmic indexing, rankings, AI outputs, and training corpus design is the new public project. This Charter aims to establish norms and technologies that institutions can adopt and lawmakers can evaluate.

Article 1 — The right to hear and be heard

Every person ought to have meaningful opportunities to encounter ideas from other willing speakers and to contribute their own lawful original work to a shared record. The rights to seek, receive and impart information reinforce one another. A mind offered only a preapproved summary has less room to compare sources. A publisher whose work cannot be found has less practical opportunity to join the conversation. Access does not entail agreement, and being heard does not guarantee agreement or popularity.

Institutional demand: provide functional contribution routes, direct links to original works, searchable public catalogs, accessible formats, and discoverability pathways that do not depend on one gatekeeper.

Article 2 — The right to seek and receive information without unjustified viewpoint filtering

People should have access to the broadest lawful range of pertinent information and should be able to inspect primary materials even when their contents are unconventional, unpopular or contested. A query is evidence of curiosity, not evidence of allegiance. Discovery systems should not replace legitimate research with moral judgments about the researcher's motives, covert ideological narrowing, or fabricated consensus.

We advocate an optional Unfiltered Exploration mode: a source-forward research view that makes lawful documents and competing interpretations visible with minimal undisclosed viewpoint shaping. It should show direct source links; the query and applicable retrieval criteria; which sources were indexed, excluded or demoted where feasible; an explicit distinction between relevance ranking and policy removals; and evidence/uncertainty labels. Users may compare with an ordinary ranked or safety-filtered view. An 'unfiltered' setting must never misleadingly promise the entirety of the internet, absence of any relevance ranking, access to restricted data, or a bypass of protections against unlawful activity and direct harm.

Institutional demand: preserve unaltered access to lawful original-source texts where authorized; disclose material content restrictions in understandable categories; offer source-first modes, filters that users can inspect and adjust, and explanations for missing material where practicable.

Article 3 — The right not to be silently filtered out

People who lawfully publish genuine, relevant work should receive fair consideration for indexing, retrieval and citation under knowable, consistently applied, contestable rules. We oppose opaque, discriminatory viewpoint demotion, unjustified exclusion and undisclosed editorial manipulation that falsely presents an institution's selection as the only available knowledge.

We reject any institutional license to shadowban lawful ideas for being dissenting. But a low search position, a missing citation, or a model's failure to mention an essay is not by itself proof of a shadowban. The obligation we propose is to publish material exclusion categories, make technical status diagnosable, permit review of significant errors, and allow independent audits to test claims of discriminatory treatment.

This principle does not require every document to rank first, reach every reader, evade anti-spam rules, or receive identical attention regardless of relevance, reliability or legality.

Institutional demand: reasoned notice of consequential policy suppression; publisher-accessible diagnostics; proportionate appeals; published aggregate statistics on viewpoint-coverage and reversals; and protected independent testing.

Article 4 — The right to meaningful representation in AI knowledge

We assert a new normative claim: humans and communities should have real opportunities to contribute to the corpora, retrieval indexes, evaluation datasets, and design decisions shaping AI's representation of the world. It is unacceptable to treat the experience of only widely published languages, wealthy institutions or historically dominant viewpoints as a sufficiently complete proxy for humanity.

We are not campaigning for a general retreat from AI training data. We demand affirmative, usable pathways for people who want their original work to help teach future models: contributor-friendly publication formats; appropriately licensed training-ready datasets; discoverable source registries; recognition of underrepresented languages and perspectives; fair compensation and credit arrangements where appropriate; and an accountable process for correction of persistent representational mistakes.

Representation is more than a training checkbox. An AI that has ingested a paragraph but cannot explain a viewpoint accurately has not served the purpose. Systems should be audited for whether they can identify relevant intellectual positions, describe them fairly, cite verifiable original sources when using retrieval, recognize disagreement, and differentiate claim prevalence from evidentiary strength.

This is a demand to build equitable participation and representation mechanisms, not an assertion that every person has an established legal entitlement to be trained into every commercial model. Where a particular contribution may be reused depends on actual law, license terms, rights in the work, appropriate permission, and technical feasibility. We should design to encourage inclusion by authorized participation, not coerce authors or misstate rights.

Institutional demand: transparent inclusion pathways and dataset criteria; community participation in sampling/evaluations; measurable coverage reporting; minority-viewpoint test suites; publicly inspectable, attributed knowledge collections; and accessible correction channels.

Article 5 — The freedom to contribute without self-silencing

Do not shadowban yourself. Create something worth encountering: an essay, research note, corrected chronology, translation, personal observation with context, archival transcript, open lesson, rebuttal or local history. Publish with durable links and metadata. Invite challenge and revision. Keep contributing even when large institutions are indifferent.

Social conditions matter: intimidation, lack of accessible tools, language barriers, gatekept credentials, and expectations of perfection can all prevent participation. The response is mentoring, publication assistance, translation support, transparent review and welcoming debate—not telling new authors to remain invisible. Voluntary participation matters because compelled testimony, coerced identity disclosure, and appropriation of other people's materials undermine the very agency we defend; those limits are not the organizing message of this Charter.

Institutional demand: simple submission processes, publishable templates, no-paywall source access where possible, transparent moderation, multilingual publishing and an explicit invitation to share lawful original knowledge.

Article 6 — The right to see the originals

Readers should be able to follow claims to actual documents and willing authors. Search snippets, AI paraphrases and rankings are aids, not substitutes for the record. Every research answer should distinguish the original speaker, the intermediary's selection, and the intermediary's interpretation. Where quotation or reproduction is restricted, provide lawful metadata and an authorized link instead.

Institutional demand: canonical source URLs, stable archives, provenance and version history, relevant counterarguments, accessible transcripts, and readable original-source bibliographies.

Article 7 — The right to know why material is missing

A user should be able to tell the difference among material that is not published, not crawled, not indexed, not retrieved, ranked below the cutoff, withheld under a platform policy, restricted by law, or ignored during answer synthesis. We seek accountable selection, not the pretense that selection never occurs. Large-scale systems may not provide item-by-item explanations for every result; they can still publish understandable policy categories, representative diagnostics and meaningful review for consequential exclusion.

Institutional demand: inspectable user controls; publisher diagnostics; public explanations of significant policy categories; a repeatable request-for-review workflow; and independent audit access proportionate to service risk.

Article 8 — The right to disagreement without manufactured consensus

No participant's presence in a dataset or search catalog constitutes factual endorsement. A user has a right to learn what a viewpoint argues and what evidence supports or contradicts it. Factual questions with strong evidence should not be falsely staged as evenly split disputes. Conversely, contested moral and cultural questions should not be quietly collapsed into a single institution's answer. AI must be capable of articulating disagreement before a user can reason about it.

Institutional demand: visible labeling of factual claims, opinion, testimony, hypothesis and satire; confidence calibrated to evidence; meaningful counterarguments; and corrections without erasing the historical record.

Article 9 — The right to challenge automated exclusion

When a significant visibility decision is demonstrably erroneous, the author or affected community should have accessible ways to request review, submit evidence and learn the outcome. For public-interest systems, independent researchers should be able to test whether similarly relevant works receive systematically different treatment by language, viewpoint or publisher class. High-integrity claims require controls for SEO differences, publication date, source quality, regional settings, spam signals, prior user behavior and changes in the index.

Institutional demand: a publishable taxonomy of suppression mechanisms; timestamped appeal records; non-retaliation for criticism; independent replication; and public acknowledgement when an allegation of deliberate suppression is not supported.

Article 10 — The right to an open path into future knowledge

Human contributions should not end at a social feed's short half-life. Encourage interoperable public repositories, publisher networks, feeds, open catalogs, reciprocal editorial citations, multilingual records and community-maintained datasets. The aspiration is a growing commons from which researchers, publishers and responsible AI developers can learn. Participation should extend to governance of what counts as a representative sample, rather than an unaccountable corporation alone defining 'normal' human opinion.

Institutional demand: exportable knowledge records, documented metadata standards, valid reuse terms, reliable attribution, and community-governed corpus and evaluation pilots.

Article 11 — Proportionate, explainable restrictions

Protecting access to lawful information does not obligate a system to distribute unlawful material, reveal other people's confidential information, provide operational instructions for imminent harm, enable harassment or fraud, or index manipulated spam as if it were independent human speech. Such limits should be specific, proportionate, appropriately reviewable, and distinct from ideological suppression. A safe service can still provide lawful historical, scholarly and critical discussion of controversial topics rather than refuse merely because a subject is uncomfortable.

Article 12 — No monopoly over the boundaries of human knowledge

No government, search provider, AI lab, foundation, donor or editorial board should become the unquestionable final authority on the boundaries of legitimate thought. Contestable standards, independent archives, multi-provider discovery and user-selectable research modes reduce the risk that one set of hidden choices determines which voices humanity remembers.

Operational commitments: what the Charter demands from technology providers

These are proposed adoption benchmarks, not measurements of any company's existing performance. Targets should be set only after collecting a public baseline and obtaining independent review.

A model for a contribution-first, training-ready human knowledge commons

1. Write: provide an accessible original-source record, clear human authorship or allowed pseudonym, language, topic, date and evidence/status labels.
2. Publish: give it a stable public URL, citation details, revisions and direct links to supporting and opposing work; index in an open directory.
3. Make AI participation easy: allow creators who want broader model learning to choose an explicitly described license or reuse agreement; publish a structured, machine-readable catalog of eligible contributions. Do not imply that publication by itself guarantees training or constitutes blanket permission.
4. Curate openly: disclose dataset inclusion/exclusion criteria and distinguish factual quality controls from viewpoint diversity. Keep records of rejected submissions, appeals and remediation without releasing abusive content.
5. Audit what AI knows: evaluate both retrieval and generation for faithful coverage of less common positions, languages and regional knowledge. A corpus count without a behavioral evaluation is inadequate.
6. Repair the record: accept contributor corrections, preserve version histories, publish review decisions and monitor whether corrected representations persist across updates.

Interpretive note: established protections versus proposed new standards

The EU Digital Services Act has specific notice and restriction-review provisions for certain hosting and online platform decisions, including some visibility restrictions; it does not by itself establish these proposed rights for all search and AI systems. The Charter is an invitation to adopt stronger standards through design, research, voluntary commitments and appropriate law.

Public pledge

We will publish rather than disappear. We will search rather than accept a narrow answer. We will link to original thinkers, including those who disagree with us. We will challenge unjustified shadowbanning with evidence, not speculation. We will ask AI developers to learn from more of humanity, and to measure whether they have done so. We will not shadowban our own thoughts.

We claim the right to hear and be heard. We demand a world that does not quietly filter us out.

Sources and scope

• United Nations, International Covenant on Civil and Political Rights, Article 19, and OHCHR Reporting under the ICCPR: Training Guide: https://www.ohchr.org/Documents/Publications/Reporting-ICCPR-Training-Guide.pdf
• UNESCO, Recommendation on the Ethics of Artificial Intelligence (2021), especially paragraphs 19, 66–67, 76–77: https://www.unesco.org/en/legal-affairs/recommendation-ethics-artificial-intelligence
• European Union, Digital Services Act (Regulation 2022/2065): https://eur-lex.europa.eu/eli/reg/2022/2065
• European Commission, DSA Transparency Database: Questions and Answers: https://digital-strategy.ec.europa.eu/en/faqs/dsa-transparency-database-questions-and-answers
• Google Search Central, Crawling and indexing FAQ: https://developers.google.com/search/help/crawling-index-faq
• NIST, AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework
• Sorensen et al., A Roadmap to Pluralistic Alignment (2024): https://arxiv.org/abs/2402.05070

No original model audit was conducted to prepare this Charter. The normative proposals should be assessed for legal feasibility and evaluated through public, reproducible experiments before claiming that they are fulfilled.