A bounded research collection

Ten public routes connect rights theory, institutional case records, legal status, cross-record comparison, draft governance, and a lawful public-process toolkit.

AI can govern through inference

Institutions do not need literal mind reading to affect a person through predictions, classifications, hidden ranking, or persistent profile changes.

Some protections and limitations are documented

Official records protect freedom of thought and neural data in specified contexts; research also limits simplistic emotion-inference claims.

No uniform global legal rule

Cognitive liberty remains an umbrella concept whose legal scope, remedies, and application to indirect inference vary by jurisdiction and technology.

10Research and toolkit pages
10Cross-cutting principles
5Official legal-status records
24Public and protected source records

Research architecture

Ten routes from rights theory to evidence comparison

Each page identifies established material, uncertainty, source class, and the boundary between evidence and proposal.

Cognitive Liberty Governance Casebook

Eleven source-aware records comparing law, regulatory action, automated adjudication, platform review, employment assessment, social protection, AI companionship, and private governance without collapsing unlike evidence into one narrative.

Open research page

Cross-source synthesis

Ten recurring governance principles

These are analytical syntheses drawn from official records, research, and attributed policy proposals. They are not presented as a single enacted bill of rights.

  1. Thought is not conduct

    Private inquiry, doubt, emotion, imagination, and unmanifested belief are analytically distinct from outward threats, fraud, coercion, stalking, discrimination, or violence.

  2. Inference is not proof

    A probabilistic model output about emotion, intent, attention, honesty, or risk is not the person’s internal state and should not be treated as conclusive by itself.

  3. Prediction is not a verdict

    Forecasts and risk scores require uncertainty, proportionality, human review, and evidence of outward conduct before high-impact consequences.

  4. Mental and neural data are sensitive

    Neural, biological, behavioral, conversational, and biometric traces can reveal or infer intimate states and require purpose limitation and strong privacy controls.

  5. Influence and governance should be visible

    People should be able to distinguish removal, restriction, demotion, ranking, labeling, reframing, refusal, and profile or memory changes.

  6. High-impact decisions need reasons and appeal

    Consequential automated decisions should preserve the record, identify the action and governing rule, and provide meaningful human review.

  7. People can be more than their archive

    Historical data, abandoned beliefs, past crises, and stale inferences should not become permanent algorithmic identity without temporal context and correction.

  8. Persona and memory changes need provenance

    Material transformations to a stored persona, memory, or profile should be visible, attributable, reversible where feasible, and separated from the preserved source record.

  9. The right to leave matters

    Systems should not engineer dependency, punish disengagement, or make access to identity and records contingent on continued participation.

  10. Human responsibility remains

    Automated systems do not absorb the legal, ethical, or editorial responsibility of the people and institutions that design, deploy, or rely on them.

Right-to-know taxonomy

Different actions require different records

Visible removal, hidden demotion, generated reframing, risk scoring, and memory changes should not be collapsed into one generic concept of moderation.

Forms of AI-mediated governance and typical notice
ActionWhat changesTypical visibility
RemovalContent is deleted, blocked, deindexed, or made unavailable.Usually visible, but reason quality varies.
RestrictionContent remains but access, geography, age, account functions, or searchability is limited.Sometimes visible.
DemotionRanking or recommendation systems reduce discoverability without deleting the material.Often not visible.
LabelingA warning, context label, trust cue, or monetization state changes how material is received.Visible, while downstream distribution effects may not be.
ReframingA summary, answer, or generated overview selects what counts as the gist and what is omitted.The output is visible; omitted alternatives usually are not.
Risk scoringA model classifies an account, person, claim, or activity for extra scrutiny or restricted access.Frequently opaque.
Memory or profile changeA persistent user record, persona, preference, or inferred identity is updated, combined, or removed.May be only partly visible.

Named external governance source

UAIX Cognitive Liberty Charter Draft

A public UAIX governance and interoperability draft concerning lawful inquiry, adult agency, persona integrity, transparent boundaries, mental privacy, least-restrictive safeguards, review, appeal, and public accountability.

Methodology

Source class stays attached to the claim

Seven submitted research packages were preserved privately. Public text separates their synthesis from official records and from the external UAIX draft.

Primary or official record
A statute, regulation, court record, official report, or institutional source.
Enacted law
A measure shown by an official legislative source as approved or in force; scope and interpretation remain jurisdiction-specific.
Research finding
An empirical or scholarly result that may be limited by design, sample, context, or replication.
Legal interpretation
An analysis of rights or obligations, not a binding holding unless a court or competent authority says otherwise.
Policy proposal
A proposed rule, safeguard, or governance model that has not automatically become law.
Advocacy framework
A manifesto, pledge, speech, or normative declaration attributed to its source.
Draft governance
A public draft that expresses intended governance practice but is not law, certification, or runtime authority.
Unknown or requires review
A claim whose present legal status, implementation, prevalence, or effect has not been established.