AI PSYOPS · FIELD REALISM · WIP.56
A defensive authority-review layer for deciding what an incident, model, law, appeal, forecast, or authentication signal actually establishes.
The twelve-category taxonomy describes how AI can intersect with influence. Field realism follows the evidence farther: from source identity to authority, from artifact to effect, from appeal to downstream repair, from sensor to construct validity, and from predictive output to the institutional action taken on it.
A CROSS-CUTTING EVIDENCE LAYER
WIP.56 retains the twelve exact owner-supplied report records and their thirteen source instances. It adds twenty-two primary or authoritative source records and thirteen claim-specific reviews. The additional predictive-population deployment report is explicitly repository-authored because no new dedicated owner report on that topic was supplied in the WIP.55 intake.
What this changes: public wording now states the source relationship, supported finding, unsupported stronger inference, competing explanations, measurement and generalization limits, privacy and affected-person boundaries, currentness, correction trigger, and reopening trigger.
What this does not change: source review is not a legal opinion, clinical diagnosis, scientific certification, intelligence attribution, affected-community endorsement, accessibility certification, publication approval, production authority, or permission to conduct psychological operations.
SIX REALISM GATES
Each gate prevents a common analytical collapse. A source can pass one gate and still leave every later stage unresolved.
An artifact can exist without proving coordination, actor identity, sponsorship, intent, recipient belief, behavior change, or strategic effect.
A reversal or restored item can stop an immediate error without correcting copied data, derived scores, rankings, vendor records, lost income, reputation, or repeated harm.
A facial movement, vocal feature, gaze pattern, physiological signal, or interaction trace is not direct knowledge of emotion, deception, loyalty, vulnerability, or intent.
One enacted rule, judgment, settlement, guidance document, or proposal does not become a universal cognitive-liberty code.
An attributable first-person or institutional record can document harm without establishing population prevalence, one exclusive cause, or community endorsement.
A generated face, detector score, linguistic pattern, or synchronized post is not enough by itself to establish a coordinated network, actor, sponsor, intent, or effect.
TWENTY-STAGE EFFECT ATTRIBUTION
Distribution metrics, psychological outcomes, operational outcomes, and strategic effects are separate findings. A sponsor’s objective is not a measured effect.
Publication rule: output does not establish distribution; reach does not establish exposure; exposure does not establish credibility; credibility does not establish belief; behavior does not automatically establish operational or strategic effect.
THIRTEEN CLAIM-SPECIFIC REVIEWS
Open a review to see the source relationship, alternatives, measurement limits, privacy boundary, currentness, and the evidence that would reopen the assessment.
A successful appeal, reversal, settlement, or restored account can stop an immediate error without repairing every copied record, derived score, ranking signal, vendor system, financial loss, reputational consequence, or psychological harm.
Remedy must be decomposed into notice, access, authoritative review, reversal, restoration, data correction, downstream propagation, compensation, repeat prevention, and unresolved harm.
That every appealed system fails downstream repair or that every reversal is ineffective.
Public decisions rarely disclose all internal ranking, feature-store, vendor, cache, and backup changes.
A platform-content appeal and a public-benefits inquiry are different legal and technical systems.
Do not publish personal records or infer private harms beyond attributable evidence.
Published testimony and institutional findings can establish experienced harms without making every account representative.
Correction trigger: A source shows complete downstream correction and remediation for a cited case.
Reopening trigger: New audit, compensation, propagation, or repeat-error data becomes available.
Deleting a source record or blocking an output does not by itself remove the record’s learned influence from model weights, embeddings, features, memories, caches, backups, or downstream recipients.
Source deletion, suppression, output blocking, retraining, and influence removal are distinct interventions.
That exact unlearning is feasible for every model or that a specific implementation satisfies law.
Verification requires comparing model behavior or parameters against an appropriate counterfactual and tracing downstream systems.
Methods for sharded supervised models do not automatically transfer to foundation models, reinforcement learning, or proprietary multi-vendor systems.
Verification should avoid re-exposing the very personal data being removed.
A person’s right to correction does not imply that every probabilistic inference can be labeled factually true or false without context.
Correction trigger: NIST or a recognized standard adopts a materially different definition.
Reopening trigger: Validated exact-unlearning benchmarks for large deployed models become available.
A system may detect a facial movement, voice feature, gaze pattern, physiological signal, or interaction trace without validly determining a person’s emotion, deception, loyalty, attention, vulnerability, or intent.
Sensor and feature accuracy must be separated from label quality, construct validity, calibration, generalization, base rates, subgroup burden, and institutional action.
That no behavioral or physiological signal has any predictive utility for any narrow purpose.
Ground truth for latent mental constructs is often contested, culturally dependent, and affected by context.
Facial-expression evidence cannot be generalized automatically to every modality or population.
Do not treat inferred states as private facts or publish individualized mental-health conclusions.
Subgroup performance, disability, neurodiversity, and cultural expression require separate evaluation.
Correction trigger: Large replicated field studies establish robust construct validity for a specified system and use.
Reopening trigger: New scientific consensus or regulatory guidance materially changes the boundary.
Current protections for neural data, emotion inference, manipulation, automated decisions, and mental privacy are distributed across jurisdictions and legal instruments; no single universal cognitive-liberty code governs every system.
Legal status must separate proposal, enactment, entry into force, application, guidance, complaint, enforcement, settlement, judgment, amendment, repeal, and supersession.
Legal advice, universal application, or uniform enforcement.
Public legal status may lag implementation, enforcement, or litigation.
EU and Colorado rules cannot be universalized to other jurisdictions.
Legal summaries should not expose complainants or infer protected traits.
Formal rights may remain inaccessible because of cost, language, disability, retaliation, or standing barriers.
Correction trigger: A law is amended, repealed, corrected, or authoritatively interpreted.
Reopening trigger: Implementing rules, enforcement, or appellate decisions become available.
Removal, restriction, search exclusion, recommendation exclusion, downranking, labeling, monetization change, account penalties, synthesized reframing, and profile changes are different interventions with different notice, evidence, appeal, restoration, and downstream effects.
A content item can be restored without evidence that reach, recommendation, strikes, monetization, or copied signals were repaired.
That every traffic decline is suppression or every ranking change is censorship.
Platforms rarely reveal counterfactual distribution, recommendation coefficients, or all internal enforcement signals.
One platform decision does not describe every platform or AI assistant.
Appeal transparency should not expose reporter identities, private safety data, or vulnerable users.
Creator income and audience effects require attributable evidence rather than inferred reach loss.
Correction trigger: Platform records show that a named action or outcome was misclassified.
Reopening trigger: Audit data reveals downstream ranking, strike, monetization, or profile changes.
Attributed first-person, representative, institutional, aggregated, anonymized, and derivative evidence can document serious algorithmic harms, but none should be treated as automatically representative of every affected person or as proof of one exclusive causal mechanism.
Affected-person evidence belongs in governance analysis with consent, privacy, retaliation, selection, nonresponse, and representativeness limits.
Population prevalence from a case collection or an exclusive causal chain for every reported harm.
Nonresponse, survivorship, legal selection, and inaccessible records limit inference.
One program, community, or jurisdiction cannot represent all systems.
Do not identify private people, publish sensitive records, or infer diagnosis, immigration status, disability, or trauma.
No project-run interviews, consent, or community endorsement are claimed in WIP.56.
Correction trigger: A cited account is corrected, withdrawn, or shown to be misrepresented.
Reopening trigger: Authorized community research, representative surveys, or new official records become available.
The FTC’s companion-chatbot inquiry establishes an official demand for information about safety, children and teens, disclosures, monetization, and data practices; it is not a completed finding that any product caused a psychiatric condition.
Companion-system safety requires product-design, age, disclosure, data, engagement, crisis, and dependency review.
Prevalence, diagnosis, causation, negligence, or liability.
Case reports, media accounts, and complaints cannot establish population prevalence or comparative risk alone.
Different products, model versions, ages, and usage patterns have different risk profiles.
Do not diagnose identifiable users or publish private chats without authorization.
Families and users should be represented through attributable, consent-aware records with withdrawal and retaliation safeguards.
Correction trigger: FTC publishes findings or corrects the inquiry scope.
Reopening trigger: Peer-reviewed longitudinal evidence, final enforcement, or adjudicated product records become available.
The New Hampshire Biden robocall is documented as an AI-generated impersonation, spoofed distribution event, and enforcement matter; the available authority does not establish how many recipients believed it or whether voting behavior changed.
Incident analysis must separate existence, content, coordination, identity, sponsorship, intent, output, distribution, availability, reach, exposure, attention, recall, comprehension, credibility, belief, intention, behavior, operational outcome, and strategic effect.
Recipient belief, turnout change, electoral outcome, or broad AI persuasion effectiveness.
Recipient-level exposure, belief, and behavior data are not available in the cited enforcement record.
A targeted robocall cannot represent all synthetic media or election contexts.
Do not publish recipient phone numbers or private voter data.
Recipient experience is not inferred without attributable testimony or research.
Correction trigger: FCC or a court changes the operator, scope, or final disposition.
Reopening trigger: Survey, telecom, or election research measures exposure, belief, or behavior.
Synthetic profile imagery or an AI-content detector score can be one clue, but coordinated inauthentic behavior should be assessed through network, chronology, infrastructure, role, and platform-enforcement evidence; a generated face alone does not identify a sponsor or prove effect.
Content generation, coordination, actor identity, sponsorship, intent, and effect are separate claims.
That pseudonymous users, non-native writers, activists, role-players, or neurodivergent users are bots because of linguistic style or imagery.
Platform access and private telemetry create verification asymmetry; public researchers may lack decisive metadata.
Detection signals vary by platform, language, community, and adversary adaptation.
Avoid de-anonymizing lawful pseudonymous users or publishing sensitive network identifiers.
False-positive burden must be evaluated for marginalized, anonymous, and emergency-response communities.
Correction trigger: A platform retracts or materially changes a network attribution.
Reopening trigger: Independent forensic, legal, or financial evidence establishes actor or sponsor identity.
Official Ukrainian records document the Zelenskyy surrender deepfake and later Zaluzhnyi deepfakes and describe rapid detection and debunking; they do not quantify continuing belief, troop-level exposure, operator identity for every artifact, or strategic effect.
Rapid live confirmation, official channels, redundancy, platform response, and correction archives are distinct crisis controls.
Precise origin, continuing belief, behavior change, operational disruption, or strategic effect.
Audience belief and frontline exposure are not fully measured in the official report.
A wartime national-leader incident differs from local fraud, elections, or private-person harassment.
Do not expose private recipients, device logs, or operational security details.
Civilian and service-member experiences require direct, consent-aware evidence.
Correction trigger: The issuing body or stronger forensic evidence changes chronology or attribution.
Reopening trigger: Platform telemetry, intelligence release, or recipient research measures exposure and effect.
Cryptographic provenance can verify a signed history or claim, but it does not prove the represented event is true, and missing credentials do not prove media is false. Experimental evidence also shows that false deepfake claims do not work uniformly across formats and contexts.
Authentication should combine provenance, source confirmation, live interaction, reverse search, forensics, channel redundancy, correction archives, and uncertainty disclosure.
That provenance guarantees truth or that deepfake denial always succeeds.
Trust-list governance, key compromise, metadata loss, and user comprehension affect real-world performance.
Survey experiments do not reproduce every high-stakes crisis or media ecosystem.
Provenance systems should minimize unnecessary device and creator identity exposure.
Authentication should not require vulnerable witnesses to reveal identities publicly.
Correction trigger: C2PA or the journal corrects the relevant records.
Reopening trigger: Field evidence on credential adoption, attack resistance, or liar’s-dividend effects becomes available.
A defensible cognitive-security mandate can focus on concealed foreign direction, artificial amplification, impersonation, technical tampering, and coordinated inauthentic behavior while preserving lawful domestic belief, dissent, anonymity, journalism, and political opposition.
Mandates should be specific, legally bounded, transparent, independently overseen, correction-capable, and separated from content or belief policing.
That every institutional classification is correct or that the French model is universally transferable.
Effectiveness and false-positive rates require independent oversight and published outcome data.
Constitutional, intelligence, platform, and administrative law differs across countries.
Open-source analysis should not become general monitoring of lawful domestic communities.
Targets of erroneous classification need notice, correction, and remedy where disclosure is legally possible.
Correction trigger: VIGINUM changes its mandate or an oversight body identifies mischaracterization.
Reopening trigger: Independent audits, litigation, parliamentary review, or affected-community evidence becomes available.
Population-level forecasts used for planning must be separated from person- or family-level risk scores used to route, investigate, screen, deny, or intensify intervention. Accuracy, authority, rights burden, feedback loops, and remedy differ substantially between these system classes.
Every predictive system should disclose unit, target, horizon, data, model role, threshold, action, evaluation, false-positive and false-negative burden, subgroup effects, human authority, feedback loops, appeal, and current status.
That all forecasting is harmful, that all person-level scoring is unlawful, or that a model’s operator description proves independent validity.
Operator transparency records often omit independent calibration, subgroup error, and counterfactual outcomes.
Health, humanitarian, conflict, visa, child-welfare, fraud, and policing systems have different purposes and legal bases.
Population planning should minimize reidentification; person-level systems require heightened data minimization and access control.
No system should infer endorsement from affected communities; direct and representative evidence remains necessary.
Correction trigger: An operator or court changes a system’s purpose, authority, status, or outcome.
Reopening trigger: Independent evaluation, litigation, audit, or affected-person evidence becomes available.
REMEDY AND SCIENTIFIC VALIDITY
A nominal appeal is not complete repair. A high classification score is not proof that a system measures the claimed psychological construct.
REPOSITORY-AUTHORED BOUNDED RESEARCH
No dedicated owner-supplied predictive-deployment report was present in the WIP.55 intake. WIP.56 fills that gap with a clearly labeled synthesis of primary official records rather than manufacturing an owner source.
Examples: Influenza activity, forced displacement, conflict risk
Usually planning-oriented; uncertainty, resource allocation, omission, and politicized interpretation remain material.
Examples: Visa application complexity routing
Not necessarily a final decision, but can alter delay, scrutiny, workload, and automation bias.
Examples: Police risk lists, child-welfare screening, fraud-risk indications
Heightened notice, validation, subgroup burden, feedback-loop, human-authority, appeal, and downstream-correction requirements.
| Deployment | System class | Unit | Decision role | Status | Validation and remedy boundary |
|---|---|---|---|---|---|
| Chicago Strategic Subjects List and Crime and Victimization Risk ModelPolicing and public safety | PERSON_LEVEL_RISK_SCORING |
Named individuals | Person-level scoring and tiering used within policing programs. | DECOMMISSIONED_2019-11-01 | OFFICIAL_OVERSIGHT_DOCUMENTED_INDEPENDENT_CAUSAL_BENEFIT_NOT_ESTABLISHED Downstream deletion, correction, and notice to scored people are not established in the cited authority. |
| System Risk Indication (SyRI)Public-benefits and fraud risk | PERSON_OR_HOUSEHOLD_RISK_ANALYSIS |
People and households within selected areas or investigations | Risk indication supporting government investigation rather than final adjudication by itself. | LEGAL_FRAMEWORK_FOUND_INCOMPATIBLE_WITH_ARTICLE_8_IN_2020_JUDGMENT | JUDICIAL_TRANSPARENCY_AND_PROPORTIONALITY_FAILURE_DOCUMENTED_PREDICTIVE_VALIDITY_NOT_ESTABLISHED The judgment stopped the legal framework’s use in the challenged form; complete correction of all underlying records and harms is not established. |
| Allegheny Family Screening ToolChild welfare intake | FAMILY_LEVEL_RISK_SCREENING |
Children and families named in hotline referrals | High score levels can require screening in; other scores supplement human intake judgment. | CURRENT_OPERATOR_DESCRIPTION_AS_ACCESSED | OPERATOR_DOCUMENTATION_AVAILABLE_INDEPENDENT_CURRENT_VALIDITY_REVIEW_INCOMPLETE Public operator description does not establish a comprehensive affected-family appeal and downstream-correction process. |
| UK Home Office Complexity Application Routing Solution — VisitsImmigration workflow | CASE_ROUTING_AND_PRIORITIZATION |
Visitor visa applications | The official record states that the system does not make the application decision. | CURRENT_TRANSPARENCY_RECORD_AS_ACCESSED | ROLE_AND_PURPOSE_DOCUMENTED_INDEPENDENT_GROUP_PERFORMANCE_NOT_CLOSED Ordinary visa-review routes may apply, but correction of routing features and downstream influence is not established here. |
| CDC FluSightPublic-health planning | AGGREGATE_POPULATION_FORECASTING |
Population-level influenza activity over time | Forecasts inform planning rather than assigning individual risk labels or legal consequences. | ACTIVE_PROGRAM_AS_ACCESSED | SEASONAL_EVALUATION_PRACTICE_DOCUMENTED_PERFORMANCE_VARIES_BY_SEASON_AND_HORIZON Forecast correction is principally methodological and public; individual appeal is generally not the relevant remedy. |
| UNHCR forced-displacement forecastingHumanitarian planning | AGGREGATE_DISPLACEMENT_FORECASTING |
Aggregate displacement flows and humanitarian demand | Planning input; the cited record does not assign person-level sanctions or status decisions. | 2025_TECHNICAL_WORK_RECORDED | INSTITUTIONAL_TECHNICAL_WORK_DOCUMENTED_OPERATIONAL_PERFORMANCE_REVIEW_NOT_CLOSED Model revision and allocation correction are relevant; individual appeal is not the primary mechanism in the cited use. |
| European Commission JRC Dynamic Conflict Risk ModelConflict early warning | AGGREGATE_STRUCTURAL_RISK_FORECASTING |
Country- or area-level conflict risk over defined horizons | Analytical input rather than an autonomous decision or coercive intervention. | 2025_TECHNICAL_REPORT_AS_PUBLISHED | MODEL_AND_INTEGRATION_APPROACH_DOCUMENTED_OUT_OF_SAMPLE_AND_POLICY_EFFECT_REVIEW_OPEN Correction concerns model, data, and published assessment; person-level appeal is not the primary mechanism. |
Stated purpose: Estimate risk of involvement in gun violence and support intervention prioritization.
Rights and feedback-loop risk: Historical police data can encode enforcement patterns; person-level scores can intensify contact and create feedback loops.
Human authority: Institutional intervention decisions remained human, but the score structured attention and priority.
Source supports: The OIG documents the models, governance concerns, and decommissioning.
Source does not establish: Crime reduction caused by the models, complete downstream repair, or universal properties of predictive policing.
Open the primary or authoritative record
Reopening trigger: New litigation, audit, retained-score evidence, or replacement-system record.
Stated purpose: Identify elevated risk of fraud or non-compliance by linking administrative data.
Rights and feedback-loop risk: Opaque multi-source scoring can burden disadvantaged neighborhoods and make errors difficult to contest.
Human authority: Authorities selected projects and acted on indications; human involvement did not remove transparency and proportionality concerns.
Source supports: The court’s disposition and legal reasoning in the case.
Source does not establish: That every fraud-risk system is unlawful or that every person experienced the same harm.
Open the primary or authoritative record
Reopening trigger: Higher-court, legislative, compensation, or implementation records.
Stated purpose: Estimate chance of future out-of-home placement to inform screening decisions.
Rights and feedback-loop risk: Administrative data and prior intervention can reproduce institutional patterns; mandated thresholds can amplify automation bias.
Human authority: Human screeners retain roles, but high-score mandates constrain discretion.
Source supports: Purpose, score role, and operator-described workflow.
Source does not establish: Independent fairness, causal benefit, or absence of disparate burden.
Open the primary or authoritative record
Reopening trigger: New independent audit, litigation, model update, threshold change, or community evidence.
Stated purpose: Route applications according to assessed complexity for workflow management.
Rights and feedback-loop risk: Routing can still affect delay, scrutiny, workload, and automation bias even when it is not the final decision.
Human authority: Human caseworkers make final decisions.
Source supports: The operator’s system purpose, rules-based nature, and non-decision role.
Source does not establish: Fairness, calibration, or no effect on applicants.
Open the primary or authoritative record
Reopening trigger: Impact assessment, audit, updated transparency record, or litigation.
Stated purpose: Support public-health situational awareness, planning, and resource decisions.
Rights and feedback-loop risk: Lower person-level rights burden, but communication of uncertainty and resource-allocation consequences still matter.
Human authority: Public-health officials interpret forecasts alongside surveillance and operational judgment.
Source supports: Existence, planning purpose, and evaluation model.
Source does not establish: Perfect accuracy or transferability to political or security prediction.
Open the primary or authoritative record
Reopening trigger: Season-end evaluation, methodology change, or program status change.
Stated purpose: Anticipate potential forced displacement to inform preparedness and resource planning.
Rights and feedback-loop risk: Aggregate models can still misallocate scarce aid or overlook communities if uncertainty and data gaps are hidden.
Human authority: Humanitarian analysts and decision-makers integrate forecasts with contextual evidence.
Source supports: The institutional forecasting initiative and humanitarian-planning purpose.
Source does not establish: Individual movement prediction or universal performance across crises.
Open the primary or authoritative record
Reopening trigger: Operational evaluation, model-card publication, or revised technical guidance.
Stated purpose: Provide quantitative risk signals integrated with qualitative early warning.
Rights and feedback-loop risk: False alarms and omissions can affect attention and resources; politicized interpretation can exceed the model’s evidentiary scope.
Human authority: Analysts integrate quantitative output with qualitative evidence and policy judgment.
Source supports: The model architecture and its role within a mixed-method early-warning process.
Source does not establish: Certainty of conflict onset or strategic benefit from actions taken.
Open the primary or authoritative record
Reopening trigger: Performance evaluation, model revision, or public operational-use record.
CRISIS AUTHENTICATION
Authentication fails when one detector or one credential is treated as an oracle. Provenance, forensics, official confirmation, corrections, and public literacy answer different questions.
Direct interactive confirmation can rapidly rebut a recording, but the channel itself must be authenticated and available.
Pre-established official channels reduce ambiguity; channel compromise or impersonation remains possible.
Independent channels reduce single-point failure but do not guarantee audience exposure.
Can verify signed provenance claims; cannot prove the depicted event is true or make unsigned media false.
Can signal synthetic origin in supported systems; marks may be absent, stripped, or unavailable for open models.
Can identify artifacts or inconsistencies; detector confidence is not identity, sponsorship, or effect.
Can reveal older or unaltered material; novel synthetic media may have no prior source.
Preserve institutional memory; a correction record does not prove every audience received or believed it.
Can prepare audiences for known tactics; must not become generalized distrust of all evidence.
Can coordinate verification and communication; authority, oversight, privacy, and correction must be explicit.
BEHAVIOR-BASED DEMOCRATIC DEFENSE
THIRTEEN PUBLIC REPORT RECORDS
Exact owner-source identity remains separate from authority review. The thirteenth record is visibly marked as repository-authored and has zero owner source instances.
REMEDY AND DOWNSTREAM REPAIR
Examines the gap between a nominal appeal, a reversible decision, immediate restoration, downstream-data correction, compensation, repeat-error prevention, and complete repair across high-stakes automated systems.
Grounds cognitive-liberty remedy claims in the difference between stopping an error and repairing the financial, reputational, housing, employment, benefits, identity, and psychological consequences that can continue after reversal.
The report is owner supplied and not independently citation-audited in this release. A reversed decision, settlement, restored post, or reopened account is not treated as proof that every copied record, score, vendor system, or downstream consequence was repaired.
DATA LINEAGE, CORRECTION, AND DELETION
Traces a correction or deletion request through source databases, features, embeddings, recommendation profiles, agent memories, model weights, caches, backups, vendors, and data brokers.
Makes downstream repair technically concrete by distinguishing physical deletion, suppression, tombstoning, source correction, embedding regeneration, model editing, retraining, machine unlearning, retention expiry, and output-layer blocking.
The report describes technical approaches and legal tensions; it does not establish that any particular system can perfectly remove a person’s influence from every model or recipient. Concealment at the output layer is not represented as verified erasure.
SCIENTIFIC VALIDITY AND FIELD CONSEQUENCE
Separates sensor accuracy, feature extraction, label quality, construct validity, criterion validity, calibration, generalization, base-rate sensitivity, subgroup burden, and institutional action across fourteen reported deployments.
Prevents the site from presenting detection of a facial movement, vocal feature, gaze pattern, keystroke rhythm, physiological signal, or neural trace as reliable knowledge of emotion, attention, deception, loyalty, personality, dangerousness, or intent.
Owner-supplied case descriptions and citations are not independently verified here. Vendor demonstrations and laboratory classification scores remain insufficient evidence of construct validity or real-world benefit.
JURISDICTION-SPECIFIC LAW AND IMPLEMENTATION
Surveys international, supranational, national, and subnational legal instruments touching neural data, mental privacy, biometric inference, AI manipulation, automated decisions, transparency, and contestability.
Strengthens the site’s distinction between enacted text, entry into force, application date, guidance, complaint, inquiry, enforcement, settlement, judgment, appeal, amendment, repeal, and supersession.
The report contains currentness claims extending into 2026 that require official issuing-body verification before being promoted as current law. It is not legal advice, and no jurisdiction-specific rule is treated as a universal cognitive-liberty code.
VISIBILITY ACTIONS AND OBSERVED OUTCOMES
Treats removal, restriction, search exclusion, recommendation exclusion, downranking, labeling, monetization change, synthesized reframing, account penalties, personalization, and saved-profile change as distinct actions.
Adds case-oriented questions about notice, rule-specific reasons, internal and independent appeal, restoration, strikes, ranking signals, income, audience, language burden, accessibility, and unresolved harm.
A traffic decline alone is not treated as proof of covert suppression, and not every ranking decision is called censorship. Restoration of one item is not complete repair of reach, income, reputation, or future recommendation state.
AFFECTED-PERSON AND COMMUNITY EVIDENCE
Organizes published evidence involving workers, applicants, tenants, borrowers, benefits recipients, students, families, disabled and neurodivergent people, creators, journalists, minority-language communities, migrants, neurotechnology users, and AI-companion users.
Strengthens consent, privacy, retaliation, selection, nonresponse, representativeness, and lived-consequence boundaries, while distinguishing first-person, representative, institutional, aggregated, anonymized, and derivative evidence.
The project conducted no interviews and claims no community endorsement. Illustrative accounts are not prevalence estimates, and public testimony does not eliminate privacy, safety, withdrawal, or retaliation concerns.
CLINICAL SAFETY AND AUTHORITY DISPLACEMENT
Examines parasocial dependency, simulated affection, engagement optimization, sleep disruption, identity discontinuity, social stimulation versus displacement, reality testing, crisis disclosure, and the boundary between companion products and clinical systems.
Improves realism around AI as a psychological authority by distinguishing formal diagnoses from public labels, temporal association from causation, and AI as catalyst, amplifier, co-author, or object of a delusional framework.
The report is not a clinical review of an identifiable person and does not establish prevalence or causation from media cases. The public site must not diagnose users or treat the phrase “AI psychosis” as a formal diagnostic category.
INCIDENT VERIFICATION AND EFFECT ATTRIBUTION
Applies a twenty-stage evidence model to reported election, narrative, geopolitical, reputational, and synthetic-media incidents so that artifact existence, coordination, attribution, intent, distribution, exposure, belief, behavior, and strategic effect remain separate.
Provides a concrete realism discipline for the twelve-category taxonomy: an AI artifact can exist and circulate without proving actor identity, sponsorship, persuasion, behavior change, operational outcome, or strategic effect.
Incident tables contain owner-supplied judgments and some assertive causal language that require independent source review. Virality, views, impressions, media coverage, or election outcomes are not treated as proof of unique AI-caused behavior change.
DETECTION, ATTRIBUTION, AND FALSE POSITIVES
Separates content generation from coordination, coordination from actor identity, actor identity from sponsorship, and sponsorship from intent or effect while surveying defensive evidence from chronology, networks, infrastructure, provenance, behavior, linguistics, and platform records.
Makes false-positive risk a first-class realism concern for anonymous whistleblowers, activists, non-native writers, minority-language communities, neurodivergent users, role-players, and emergency-response communities.
No single linguistic, temporal, image, or detector signal proves a synthetic persona or coordinated influence operation. Public content must not expose operational evasion or collection procedures.
CRISIS AUTHENTICATION AND INCIDENT RESPONSE
Reconstructs fifteen reported synthetic-media crises and compares proactive provenance, watermarking, forensic analysis, reverse search, official confirmation, cross-channel redundancy, pre-bunking, and rapid-response practices.
Grounds deepfake analysis in timing, first appearance, verification, official response, platform action, media correction, continuing belief, reuse, legal action, and unresolved uncertainty rather than technical realism alone.
The report is owner supplied and contains incident-level claims that require source-by-source verification. Fast correction does not prove zero exposure or zero belief, and technical anomalies do not alone establish origin or sponsor.
AUTHENTICATION INFRASTRUCTURE AND EPISTEMIC RESILIENCE
Compares cryptographic signing, content credentials, watermarking, forensic analysis, reverse search, live confirmation, correction archives, official channels, cross-channel redundancy, pre-bunking, and rapid-response teams.
Treats authenticity as an institutional system rather than a detector score and gives equal attention to the liar’s dividend: authentic evidence can be falsely dismissed as synthetic.
No provenance or detector system is represented as universal, unbreakable, or sufficient on its own. Absence of credentials is not proof of falsity, and presence of credentials requires trust-chain and custody review.
BEHAVIOR-BASED DEMOCRATIC DEFENSE
Compares institutional approaches to foreign-influence transparency, authentication, platform reporting, independent journalism, researcher access, civic verification, election communication, provenance, whistleblower protection, and oversight.
Centers the distinction between defending against concealed behavior and policing lawful belief, dissent, interpretation, or political identity, with attention to mission creep, politicization, unsupported attribution, and correction failures.
Case-study claims and current institutional status require issuing-body and jurisdiction-specific review. The report does not authorize domestic surveillance, censorship, belief scoring, or unsupported foreign attribution.
REPOSITORY-AUTHORED BOUNDED RESEARCH
Compares seven primary or authoritative records spanning person-level police and child-welfare scores, fraud-risk analysis, visa workflow routing, public-health forecasting, humanitarian displacement forecasting, and conflict early warning.
Fills the WIP.55 intake gap without fabricating an owner report. It separates aggregate forecasting from case routing and person-level scoring, then records decision role, human authority, validation status, feedback-loop risk, remedy, and current status for each deployment.
This is repository-authored bounded research, not an owner-supplied report. Operator descriptions establish purpose and workflow but do not independently prove fairness, validity, benefit, or complete remedy. No specialist disposition or affected-community endorsement is claimed.
SOURCE AND REVIEW INTEGRITY
Correction standard: identify the stable report, claim, deployment, or source ID; specify the disputed wording; provide the strongest available primary or authoritative record; and state whether the correction changes source identity, currentness, interpretation, attribution, effect, remedy, or rights analysis.