Lawful public-process toolkit
Cognitive Liberty Reader Rights Toolkit
Practical, non-legal tools for documenting automated governance, asking for notice, preserving records, evaluating consent, and preparing a human-review request.
Scope
This toolkit helps a reader document what happened and request process. It does not diagnose shadowbanning, prove discrimination, guarantee reinstatement, or replace legal advice.
Public significance
Invisible or automated decisions are difficult to challenge when the original, notice, metrics, policy version, and decision state have not been preserved.
Current record
Notice, preserved records, reasons, correction, and human review are recurring governance safeguards across risk-management and civil-society frameworks.
Evidence boundary
A platform, employer, school, agency, or court may not be legally required to provide every item in this toolkit. Available rights depend on jurisdiction and context.
Evidence-led synthesis
Key findings
- Start by identifying the exact action: removal, restriction, demotion, label, summary, score, refusal, or memory change.
- Preserve screenshots, timestamps, URLs, notices, account-status pages, output versions, and relevant metrics.
- Ask whether a decision was automated, human, or mixed and what materially influenced it.
- Request qualified human review when consequences affect livelihood, education, benefits, healthcare, identity, or access.
Limitations
What this page does not establish
- The toolkit is informational and not legal advice.
- Do not use it to evade safety systems, harass reviewers, expose private data, or make unsupported public accusations.
- A visibility change can have ordinary explanations and is not conclusive proof of censorship.
Governance safeguards
Rights-preserving controls
- Preserve evidence without collecting unrelated personal data.
- Use precise, non-accusatory language and separate observation from inference.
- Ask for the applicable policy and version.
- Request correction of inaccurate data and repair of consequential state where available.
- Use official complaint, records, ombuds, regulator, or court channels appropriate to the jurisdiction.
Research gaps
Questions the current record cannot settle
- Standardized reason codes across platforms and public agencies.
- Portable records of ranking, summary, and profile changes.
- Independent dispute resolution for high-impact AI decisions.
Reader utility
Document first, infer carefully, request process
These tools support lawful documentation and review. They do not prove censorship or create legal rights.
Diagnostic questions
- What exact action occurred: removal, restriction, demotion, labeling, ranking change, refusal, risk score, summary change, or memory/profile update?
- What evidence shows a system action rather than ordinary audience, market, or contextual change?
- Was the decision automated, human, or mixed?
- What rule, policy version, jurisdiction, and data materially influenced it?
- Did it affect visibility, searchability, revenue, account standing, access, or stored personalization?
- Is a human-review path available, and what deadline applies?
- Can the original state be exported or preserved before it changes?
Preservation checklist
- Capture the original content, prompt, output, profile, or decision screen.
- Record exact dates, times, URLs, account identifiers, and device or locale context where relevant.
- Save notices, reason codes, policy links, appeal IDs, and correspondence.
- Preserve before-and-after visibility or account-state metrics without publishing private information.
- Keep a clear distinction between direct observation, platform statement, inference, and allegation.
Appeal preparation
- Identify the decision and its consequence.
- State the observed facts without assuming motive.
- Request the governing rule, policy version, decision method, and material inputs.
- Explain the specific error or missing context.
- Request qualified human review, correction, and repair of affected state.
- Ask for a written result and a durable case identifier.
Consent questions
- Can the person refuse without losing employment, education, healthcare, housing, benefits, or essential access?
- Is the purpose specific, understandable, and separate from secondary inference?
- Is the system scientifically valid for the stated use and population?
- Can consent be withdrawn and historical data deleted or corrected?
- Are retention, sharing, and model-training uses visible?
Model transparency notice template
Action: Identify whether the system removed, restricted, demoted, labeled, reframed, scored, refused, or changed stored profile state.
Reason: State the rule or policy version, whether the action was automated or human, and the material inputs that can safely be disclosed.
Effect: State whether visibility, searchability, recommendation eligibility, revenue, access, account standing, or stored personalization changed.
Review: Provide a durable case ID, preservation of the original state, and a qualified human-review path where the stakes require it.
Sources and limitations
Source register
External links are visitor-initiated and send no referrer. Protected research inputs remain inaccessible from the public web.
-
AI Risk Management Framework
National Institute of Standards and Technology · 2023-01-26
- Supports
- Governance, documentation, measurement, human oversight, and risk-management practices.
- Limit
- Voluntary framework unless incorporated into another authority.
-
Santa Clara Principles on Transparency and Accountability in Content Moderation
Santa Clara Principles coalition · 2021-12-01
- Supports
- Notice, explanation, and appeal concepts for moderation governance.
- Limit
- Normative framework, not binding law.
-
The Invisible Editor: AI Censorship, Algorithmic Suppression, and the Right to Know
Submitted research package · 2026-07-27
- Supports
- Taxonomy of removal, restriction, demotion, reframing, profile change, notice, and appeal.
- Limit
- Research synthesis includes time-sensitive platform and regulatory claims requiring current verification.
Preserved as protected research input; no private file path is exposed.
-
When the System Watches the Mind
Submitted research package · 2026-07-27
- Supports
- Mental-state inference, coercive consent, sector risks, and mental-privacy safeguards.
- Limit
- Interdisciplinary synthesis; examples and laws require source-by-source verification.
Preserved as protected research input; no private file path is exposed.
Connected evidence
Compare the rights lens with the AI PSYOPS taxonomy
The cognitive-liberty collection examines governance and rights boundaries. The AI PSYOPS taxonomy separately examines influence mechanisms, evidence maturity, capability limits, and defensive responses.