data/claims.csv

claim_id,case_id,claim_type,claim_text,evidence_label,claim_status,evidence_basis,evidence_scope,source_ids,counter_source_ids,notes,review_status,verified_at
clm-ro-001,ro-2024-presidential-annulment,institutional_outcome,"On 6 December 2024, Romania's Constitutional Court annulled the entire presidential electoral process.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-ro-ccr-20241206|src-ro-ccr-summary-2025,,The institutional act is certain even where its factual and legal justification is debated.,verified,2026-08-11
clm-ro-002,ro-2024-presidential-annulment,mechanism,The official record described opaque and coordinated online promotion that created unequal campaign conditions.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-ro-ccr-summary-2025|src-ro-ifes-202412,,"The sources establish that official and independent records described these conditions; they do not resolve every network component, sponsor, or causal effect.",verified,2026-08-11
clm-ro-003,ro-2024-presidential-annulment,response,"The European Commission opened formal DSA proceedings against TikTok concerning election-risk management, recommender systems, political advertising, and coordinated inauthentic manipulation.",established_evidence,supported_as_worded,supported_as_worded,institutional_action,src-ro-ec-tiktok-20241217,,Opening proceedings is not a final infringement finding.,verified,2026-08-11
clm-ro-004,ro-2024-presidential-annulment,ai_role,"Among the public sources linked to this record at the cutoff, none established generative AI as a material driver of the Romanian annulment case.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-ro-005,ro-2024-presidential-annulment,reach,"TikTok reported large-scale prevention of fake engagement and removal of election-related networks, but these platform metrics do not measure voter exposure to the specific campaign.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-ro-tiktok-integrity,,"Platform self-report; preserve distinction between prevented actions, removed accounts, impressions, and people reached.",verified,2026-08-11
clm-ro-006,ro-2024-presidential-annulment,electoral_effect,"Among the public sources linked to this record at the cutoff, none provided a credible causal estimate of votes changed by the coordinated digital campaign.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-ro-007,ro-2024-presidential-annulment,attribution,A simple claim that Russia alone directed the entire campaign is stronger than the presently coded public evidence supports.,strong_inference,partly_supported,inferential_synthesis,researcher_inference,src-ro-ccr-summary-2025|src-ro-venice-202503|src-ro-sri-csat-20241204|src-ro-snoop-pnl-20241220,,"Code official allegations, technical indicators, domestic financing, and final attribution separately.",verified,2026-08-11
clm-ro-008,ro-2024-presidential-annulment,legal_status,"The Venice Commission treated annulment as an exceptional measure requiring clear legal conditions, reliable evidence, procedural safeguards, and proportionality.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-ro-venice-202503,,"Normative legal standard, not a factual finding about every Romanian allegation.",verified,2026-08-11
clm-ro-009,ro-2025-presidential-repeat,occurrence,Romania held a repeat presidential election on 4 and 18 May 2025.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-ro-odihr-repeat-final,,,verified,2026-08-11
clm-ro-010,ro-2025-presidential-repeat,electoral_effect,The repeat-election outcome cannot be used as a causal measure of what the annulled 2024 result would have been absent manipulation.,strong_inference,partly_supported,inferential_synthesis,researcher_inference,src-ro-odihr-repeat-final,,"The electorate, candidates, campaign, information environment, and institutional context changed.",verified,2026-08-11
clm-ro-011,ro-2024-presidential-annulment,occurrence,"A declassified SRI report described an initial network of about 25,000 TikTok accounts activated intensely in the final two weeks before the first round, including 797 accounts created in 2016 that had remained largely inactive until 11 November 2024.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-ro-sri-csat-20241204,,Established as an intelligence-service report. The account count is not independently audited unique reach and does not establish the operator or voter effect.,verified,2026-08-11
clm-ro-012,ro-2024-presidential-annulment,reach,"TikTok's separately disclosed Turkey-based network and the SRI-reported 25,000-account network have different scopes and cannot be summed as one operation or one audience estimate.",established_evidence,supported_as_worded,supported_as_worded,measurement,src-ro-tiktok-integrity|src-ro-sri-csat-20241204,,"This is a source-scope correction, not an assertion that the networks were unrelated in every respect.",verified,2026-08-11
clm-ro-013,ro-2024-presidential-annulment,counterevidence,"Snoop reported that PNL financed the #EchilibrușiVerticalitate influencer campaign, while the contracted firm said its original brief had been altered, diverted, or cloned in Georgescu's favour.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-ro-snoop-pnl-20241220,,"The reporting is counterevidence to a one-actor account of all amplification. It does not establish that PNL intended to support Georgescu, nor does it disprove separate foreign-linked activity.",verified,2026-08-11
clm-ro-014,ro-2025-presidential-repeat,occurrence,"DFRLab identified at least 215 coordinated or apparently hijacked Facebook, Instagram, and TikTok accounts active from December 2024 through 6 June 2025 that first supported Călin Georgescu and later George Simion while attacking Moldova, France, and the EU.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-ro-dfrlab-repeat-20250826,,Established as a transparent public-data research finding; platform access thresholds may leave additional accounts unobserved.,verified,2026-08-11
clm-ro-015,ro-2025-presidential-repeat,reach,"The studied network produced 8,514 posts and accumulated at least 16 million views and 681,000 likes across the three platforms.",established_evidence,supported_as_worded,supported_as_worded,measurement,src-ro-dfrlab-repeat-20250826,,"These are non-unique platform metrics, not authenticated Romanian voters, belief, or persuasion.",verified,2026-08-11
clm-ro-016,ro-2025-presidential-repeat,ai_role,"The network used generative-AI profile images and descriptions, synthetic Romanian personas, and DreamFace or D-ID-style avatars alongside hijacked accounts and conventional stock footage.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-ro-dfrlab-repeat-20250826,,"AI was one component of a broader cross-platform coordination system, not a complete explanation for its reach.",verified,2026-08-11
clm-ro-017,ro-2025-presidential-repeat,attribution,The network's operator and sponsor remain publicly unidentified; pro-Russia narratives and coordinated behaviour do not by themselves establish Russian-state command.,open_question,unresolved,none,researcher_inference,src-ro-dfrlab-repeat-20250826,,"Confidence refers to the appropriately unresolved wording, not confidence that no attribution will ever be possible.",verified,2026-08-11
clm-ro-018,ro-2025-presidential-repeat,electoral_effect,"Among the public sources linked to this record at the cutoff, none estimated whether the cross-platform network changed belief, turnout, candidate preference, or the repeat-election result.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-md-001,md-2024-presidential-referendum,occurrence,Moldova held the first presidential round and EU constitutional referendum on 20 October 2024 and the presidential runoff on 3 November 2024.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-odihr-final-20250314,,,verified,2026-08-11
clm-md-002,md-2024-presidential-referendum,institutional_outcome,The EU constitutional amendment was approved by a narrow final majority and Maia Sandu won the presidential runoff.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-odihr-final-20250314,,The dataset does not infer that interference explains the margin.,verified,2026-08-11
clm-md-003,md-2024-presidential-referendum,mechanism,International observers documented active disinformation and serious concerns about illicit foreign interference affecting campaign conditions.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-odihr-prelim-20241021|src-md-odihr-final-20250314,,"Observation finding about campaign environment, not a quantified electoral effect.",verified,2026-08-11
clm-md-004,md-2024-presidential-referendum,mechanism,Moldovan police alleged that the Shor-linked network transferred tens of millions of dollars to influence voters.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-md-reuters-payments-20241024,,Established as a police allegation reported by Reuters; underlying transfers and intent require their own evidentiary assessment.,verified,2026-08-11
clm-md-005,md-2024-presidential-referendum,attribution,"Authorities and international institutions attributed major components of the operation to Russia-linked and Shor-linked actors, while Russia and Shor denied wrongdoing.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-md-odihr-final-20250314|src-md-reuters-payments-20241024,src-md-reuters-payments-20241024,Wording preserves attribution as an official assessment and the denials as counterposition.,verified,2026-08-11
clm-md-006,md-2024-presidential-referendum,ai_role,"AI-generated impersonation and synthetic content appeared in Moldova's information environment, but the wider interference operation relied heavily on non-AI mechanisms.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-eeas-fimi-202401|src-md-odihr-parliament-202602,,"Avoids treating illicit finance, bots, and coordinated networks as synonymous with generative AI.",verified,2026-08-11
clm-md-007,md-2024-presidential-referendum,mechanism,Meta reported removing a coordinated inauthentic network targeting Russian-speaking audiences in Moldova.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-md-meta-threat-reporting,,Platform self-report; independent replication is limited and gross follower counts largely reflected non-Moldovan or fake engagement.,verified,2026-08-11
clm-md-008,md-2024-presidential-referendum,counterevidence,"The 2026 ODIHR report cited a heterogeneous aggregate of more than 138,000 accounts and transactions across linked 2024–2025 investigations. It cannot be assigned to a single geographic scope, unit, unique population, or voter-reach measure.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-md-odihr-parliament-202602,,Retained only as a warning about an official-report aggregate. It is excluded from observations and from the case-level legacy reach fields.,verified,2026-08-11
clm-md-009,md-2024-presidential-referendum,behavioural_effect,"Documented payment systems provide evidence of attempted and in some cases accepted inducement, but do not establish how each recipient voted.",strong_inference,partly_supported,inferential_synthesis,researcher_inference,src-md-reuters-payments-20241024|src-md-odihr-final-20250314,,Acceptance of a payment is not identical to compliance with an instruction.,verified,2026-08-11
clm-md-010,md-2024-presidential-referendum,electoral_effect,"Among the public sources linked to this record at the cutoff, none provided a credible causal estimate of referendum or presidential votes changed by the influence operation.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-md-011,md-2024-presidential-referendum,counterevidence,"Observers also found incumbent advantage, insufficient safeguards against misuse of administrative resources, and media imbalances.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-odihr-prelim-20241021|src-md-odihr-final-20250314,,Prevents the foreign-interference frame from obscuring domestic electoral-integrity problems.,verified,2026-08-11
clm-md-012,md-2025-parliamentary-linked-network,mechanism,"The 2025 parliamentary-election observation record described inauthentic account networks, troll farms, automated bots, AI videos, disinformation, and foreign-funded vote-buying infrastructure.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-md-odihr-parliament-202602,,Evidence concerns a linked later election and is not silently back-projected into every 2024 event.,verified,2026-08-11
clm-md-013,md-2025-parliamentary-linked-network,electoral_effect,The persistence of a network across elections does not establish that AI-generated content or bots determined the 2025 outcome.,strong_inference,partly_supported,inferential_synthesis,researcher_inference,src-md-odihr-parliament-202602,,,verified,2026-08-11
clm-md-014,md-2024-presidential-referendum,reach,"Meta reported about 4,200 Facebook Page followers and 335,000 Instagram followers for the removed Moldova-focused network, but said the vast majority were outside Moldova and indicative of fake engagement; it assessed that the operation had not built an authentic audience before disruption.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-md-meta-threat-reporting,,Established as platform telemetry and assessment. Gross followers must not be reported as Moldovan voter reach.,verified,2026-08-11
clm-md-015,md-2025-parliamentary-linked-network,occurrence,"DFRLab analysed 253 inauthentic accounts in a paid InfoLeaders network linked through undercover reporting to Telegram-based recruitment, training, task assignment, and mutual amplification.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-dfrlab-infoleaders-20250924,,The platform sample and the wider number of recruited people are different denominators.,verified,2026-08-11
clm-md-016,md-2025-parliamentary-linked-network,reach,"The 253-account InfoLeaders sample produced 28,708 posts, 55,367,950 views, 2,207,894 likes, 295,608 shares, and 239,016 comments across TikTok, Facebook, and Instagram.",established_evidence,supported_as_worded,supported_as_worded,measurement,src-md-dfrlab-infoleaders-20250924,,"Platform totals are non-unique, can include repeat or inauthentic engagement, and do not measure belief or votes.",verified,2026-08-11
clm-md-017,md-2025-parliamentary-linked-network,mechanism,"Curators required local-looking participants to personalise centrally supplied messages while retaining assigned hashtags and to amplify one another, making paid coordination appear to be spontaneous peer speech.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-md-dfrlab-infoleaders-20250924,,"AI-assisted rephrasing was an enabling layer within a paid human and bot-coordinated system, not the sole mechanism.",verified,2026-08-11
clm-md-018,md-2025-parliamentary-linked-network,attribution,"Among the public sources linked to this record at the cutoff, none independently established direct Kremlin command of every InfoLeaders account or asset.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-md-019,md-2025-parliamentary-linked-network,electoral_effect,"Among the public sources linked to this record at the cutoff, none provided a causal estimate of belief, turnout, votes, seats, or the 2025 result attributable to the coded network, synthetic media, or automated accounts.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-nh-001,us-nh-2024-biden-robocall,occurrence,An AI-generated voice impersonating President Joe Biden was used in robocalls before the New Hampshire presidential primary.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fcc-kramer-nal-20240523|src-us-nhdoj-kramer-20240523|src-us-fcc-kramer-fine-20240926,,,verified,2026-08-11
clm-nh-002,us-nh-2024-biden-robocall,mechanism,The call told recipients to save their vote for November rather than participate in the primary.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fcc-kramer-nal-20240523|src-us-nhdoj-kramer-20240523,,The content has a demobilising instruction; criminal intent and legal guilt are separate questions.,verified,2026-08-11
clm-nh-003,us-nh-2024-biden-robocall,attribution,Steve Kramer admitted commissioning the AI-generated robocall.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fcc-kramer-fine-20240926|src-us-nhpr-acquittal-20250613,,Admission to creation does not equal conviction on the state charges.,verified,2026-08-11
clm-nh-004,us-nh-2024-biden-robocall,reach,"Voice Broadcasting initiated 9,581 call attempts, and Lingo records showed 3,978 completed calls bearing A-level attestations.",established_evidence,supported_as_worded,supported_as_worded,measurement,src-us-fcc-kramer-final-pdf-20240926|src-us-fcc-lingo-consent-202408,,"Calls are not equivalent to unique people reached, completed calls, or listeners.",verified,2026-08-11
clm-nh-005,us-nh-2024-biden-robocall,mechanism,Caller-ID spoofing and inaccurate attestation in the telecom chain increased the call's apparent authenticity and deliverability.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fcc-lingo-nal-pdf-20240523|src-us-fcc-lingo-settlement-20240821,,"Shows that the case is an infrastructure chain, not merely a generated audio file.",verified,2026-08-11
clm-nh-006,us-nh-2024-biden-robocall,legal_status,The FCC ruled that AI-generated voices fall within the TCPA's restriction on artificial or prerecorded voices.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fcc-ai-voice-20240208,,,verified,2026-08-11
clm-nh-007,us-nh-2024-biden-robocall,institutional_outcome,"The FCC issued a $6 million forfeiture against Kramer, while Lingo agreed separately to a $1 million civil penalty and compliance plan.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fcc-kramer-final-pdf-20240926|src-us-fcc-lingo-consent-202408,,Separate creator and provider actions.,verified,2026-08-11
clm-nh-008,us-nh-2024-biden-robocall,legal_status,A New Hampshire jury acquitted Kramer of all 22 state criminal charges in June 2025.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-nhpr-acquittal-20250613|src-us-ap-kramer-acquittal-20250613,,The acquittal must appear wherever the charging decision is discussed.,verified,2026-08-11
clm-nh-009,us-nh-2024-biden-robocall,behavioural_effect,"Among the public sources linked to this record at the cutoff, none measured how many recipients believed or acted on the robocall.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-nh-010,us-nh-2024-biden-robocall,electoral_effect,"Among the public sources linked to this record at the cutoff, none provided a credible causal estimate linking the robocall to primary turnout or vote totals.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-11
clm-nh-011,us-nh-2024-biden-robocall,institutional_outcome,"In November 2025, a federal court entered default judgment against Kramer, awarded individual monetary relief, and permanently enjoined similar voter-coercion schemes.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-lwv-kramer-civil|src-us-ap-kramer-default-202511,,"The judgment followed nonappearance and must be described as a default, not a contested merits trial.",verified,2026-08-11
clm-nh-012,us-nh-2024-biden-robocall,control_shift_hypothesis,"Hypothesis: a cloned presidential voice, selected call list, and spoofed trusted number could shift control over apparent authorship and relevance of a participation instruction toward the operator; recipient agency change is not observed.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,Analytical hypothesis linked through the pathway to factual claims; it is not direct evidence of an agency outcome.,analytical_hypothesis,2026-08-11
clm-br-001,br-2026-election-preparedness,occurrence,"Brazil's 2026 general election first round is scheduled for 4 October, with a possible runoff on 25 October.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-tse-calendar-202603,,"Prospective date, not an observed election result.",verified,2026-08-11
clm-br-002,br-2026-election-preparedness,legal_status,"TSE rules require explicit, prominent, accessible disclosure of synthetic multimedia used in electoral advertising and identification of the technology used.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-tse-res-23755-20260302|src-br-tse-ai-rules-202604,,Resolution 23.755 amends the standing electoral-advertising rule.,verified,2026-08-11
clm-br-003,br-2026-election-preparedness,legal_status,"Brazil prohibits publication or republication of new synthetic content using a candidate's or public figure's image, voice, or manifestation from 72 hours before until 24 hours after voting ends, even if labelled.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-tse-res-23755-20260302,,Time-window prohibition is distinct from the general disclosure requirement.,verified,2026-08-11
clm-br-004,br-2026-election-preparedness,legal_status,"AI application providers may not rank, recommend, suggest, prioritise, endorse, or disfavor candidates, campaigns, parties, federations, or coalitions in automated responses.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-tse-res-23755-20260302,,"A direct regulation of AI-system outputs, not only political advertising.",verified,2026-08-11
clm-br-005,br-2026-election-preparedness,response,Platforms offering political-content boosting must provide a field for advertisers to declare AI use.,established_evidence,supported_as_worded,supported_as_worded,institutional_action,src-br-tse-res-23755-20260302,,,verified,2026-08-11
clm-br-006,br-2026-election-preparedness,response,"Internet application providers must prepare measurable election-compliance plans, and the TSE set 16 August 2026 as the submission deadline.",established_evidence,supported_as_worded,supported_as_worded,institutional_action,src-br-tse-res-23755-20260302|src-br-tse-compliance-202607,,Future QA should archive submitted plans and assess measurable implementation.,verified,2026-08-11
clm-br-007,br-2026-election-preparedness,response,"By 11 August 2026, the TSE had announced platform partnerships, an AGU agreement, an information-integrity and AI monitoring council, and a voice-deepfake cooperation agreement with ElevenLabs.",established_evidence,supported_as_worded,supported_as_worded,institutional_action,src-br-tse-platforms-202607|src-br-tse-agu-20260803|src-br-tse-council-20260807|src-br-tse-elevenlabs-202608,,Institutional capacity indicators; effectiveness not yet established.,verified,2026-08-11
clm-br-008,br-2026-election-preparedness,counterevidence,"Empirical research on Brazil's 2024 municipal elections found isolated AI misuse and broad campaign experimentation, but not the predicted wave of high-impact generative-AI manipulation.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-aio-202411,,"A prior-cycle base rate, not a forecast for 2026.",verified,2026-08-11
clm-br-009,br-2026-election-preparedness,electoral_effect,"No 2026 electoral effect can be coded before voting, and observed incidents must not be converted into predictions.",open_question,unresolved,none,researcher_inference,,,Dataset guardrail rather than an empirical outcome claim.,verified,2026-08-11
clm-br-010,br-2026-bolsonaro-ai-avatar,occurrence,A synthetic image-and-voice avatar of Jair Bolsonaro appeared at the 25 July 2026 launch of Flávio Bolsonaro's presidential campaign.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-reuters-avatar-20260728,,,verified,2026-08-11
clm-br-011,br-2026-bolsonaro-ai-avatar,mechanism,The avatar explicitly disclosed that its image and voice were generated with AI and then delivered a political endorsement message.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-reuters-avatar-20260728,,A labelled synthetic political persona; not a covert deepfake in the narrow sense.,verified,2026-08-11
clm-br-012,br-2026-bolsonaro-ai-avatar,legal_status,"Opposing parties challenged the avatar in court, while Flávio Bolsonaro's lawyers argued that disclosure and the absence of mixed real footage made it lawful.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-br-reuters-avatar-20260728,src-br-reuters-avatar-20260728,Established that the competing claims were made; neither claim is treated as finally adjudicated.,verified,2026-08-11
clm-br-013,br-2026-bolsonaro-ai-avatar,legal_status,"Among the public sources linked to this record at the cutoff, no final TSE merits decision on the avatar's compliance with electoral AI rules was identified.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,needs_update,2026-08-11
clm-br-014,br-2026-bolsonaro-ai-avatar,behavioural_effect,"Among the public sources linked to this record at the cutoff, none measured whether the avatar changed voter beliefs, candidate preference, or turnout intention.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,needs_update,2026-08-11
clm-br-015,br-2026-bolsonaro-ai-avatar,legal_status,Federação Brasil da Esperança filed TSE representation RP 0601315-97.2026.6.00.0000 on 26 July 2026 alleging premature propaganda and unlawful synthetic-media use.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-br-avatar-petition-20260726,,The filing and allegations are established; the alleged violations are not.,verified,2026-08-11
clm-br-016,br-2026-bolsonaro-ai-avatar,legal_status,"Art. 9-B requires disclosure, Art. 9-B §3 bars certain simulated interlocution with a candidate or real person, and Art. 9-C §1 bars synthetic audio or video used to favor or harm a candidacy even when authorised.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-tse-res-23610-consolidated,,"The legal text is established; application to this disclosed, one-way convention avatar remains unresolved.",verified,2026-08-11
clm-br-017,br-2026-bolsonaro-ai-avatar,counterevidence,The avatar's up-front AI disclosure supports compliance with the labelling duty but does not by itself resolve the separate simulated-interlocution or pro-candidacy synthetic-media prohibitions.,strong_inference,partly_supported,inferential_synthesis,researcher_inference,src-br-reuters-avatar-20260728|src-br-tse-res-23610-consolidated,src-br-folha-avatar-defense-202607,The defence argues that disclosure and lack of deception remove it from the deepfake prohibition; no merits court had accepted or rejected that reading by the cutoff.,needs_update,2026-08-11
clm-br-018,br-2026-bolsonaro-ai-avatar,counterevidence,The 72-hour pre-election and 24-hour post-election blackout in Art. 9-B §3-A did not apply by timing to the 25 July convention.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-br-tse-res-23755-20260302|src-br-reuters-avatar-20260728,,Other provisions may still apply; retain the blackout rule for prospective monitoring only.,verified,2026-08-11
clm-br-019,br-2026-bolsonaro-ai-avatar,attribution,Jair Bolsonaro's lawyers denied that he authorised or knew of the avatar's production; his personal authorisation therefore remains unknown rather than disproved.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-br-ft-avatar-202607|src-br-reuters-avatar-20260728,,"A denial establishes the defence position, not the underlying absence of authorisation.",needs_update,2026-08-11
clm-br-020,br-2026-bolsonaro-ai-avatar,control_shift_hypothesis,"Boundary hypothesis: a disclosed synthetic surrogate could shift control over Jair Bolsonaro's public representation toward the beneficiary campaign; disclosure, unresolved authorisation, and unmeasured audience interpretation prevent a stronger conclusion.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,Analytical hypothesis linked through the pathway to factual claims; it is not direct evidence of an agency outcome.,analytical_hypothesis,2026-08-11
clm-us26-001,us-2026-midterms-preparedness,occurrence,The 2026 United States general election is scheduled for 3 November 2026.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fec-dates-2026,,,verified,2026-08-11
clm-us26-002,us-2026-midterms-preparedness,legal_status,"As of 23 June 2026, 31 states had enacted laws regulating deepfakes in political messaging, with materially different prohibition and disclosure models.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-ncsl-20260623,,"Enactment, effective date, scope, and litigation status should be tracked separately.",verified,2026-08-11
clm-us26-003,us-2026-midterms-preparedness,legal_status,The United States lacks a broad federal rule uniformly governing AI-generated political messages across the 2026 election.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-fec-ai-disposition-2024|src-us-ncsl-20260623|src-us-reuters-midterms-20260328,src-us-fcc-ai-voice-20240208,"Narrow federal rules remain relevant, especially telecom law and fraudulent misrepresentation.",verified,2026-08-11
clm-us26-004,us-2026-midterms-preparedness,ai_role,CISA assessed generative AI as more likely to amplify existing election risks than create wholly new risk categories.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-us-cisa-genai-202401,,Prior-cycle risk assessment used as a conceptual baseline.,verified,2026-08-11
clm-us26-005,us-2026-midterms-preparedness,attribution,"Prior U.S. intelligence assessments establish that foreign actors used synthetic or manipulated content in the 2024 cycle, supporting a monitored 2026 risk but not proving a new 2026 operation.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-odni-update-20241021,,Strict temporal separation between prior evidence and current incidents.,verified,2026-08-11
clm-us26-006,us-2026-midterms-preparedness,counterevidence,A platform assessment of the 2024 global cycle reported that generative-AI misinformation had limited reach on Meta services.,established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-global-meta-muted-20241203,,Interested-party assessment reported by Reuters; not generalisable to every platform or 2026.,verified,2026-08-11
clm-us26-007,us-2026-midterms-preparedness,electoral_effect,"No 2026 electoral effect can be coded before the election, and prior-cycle incident counts do not supply a forecast.",open_question,unresolved,none,researcher_inference,,,Dataset guardrail.,verified,2026-08-11
clm-us26-008,us-2026-campaign-deepfake-ads,occurrence,"On 11 March 2026, the National Republican Senatorial Committee released an AI-generated video depicting James Talarico reciting text from prior social-media posts.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-nrsc-talarico-20260311|src-us-reuters-midterms-20260328,,The source quotes the committee defending the ad.,verified,2026-08-11
clm-us26-009,us-2026-campaign-deepfake-ads,mechanism,"The Talarico ad carried an AI-generated disclosure in small, easy-to-miss text.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-reuters-midterms-20260328|src-us-berkeley-talarico-2026,,Disclosure existence and disclosure salience are distinct variables.,verified,2026-08-11
clm-us26-010,us-2026-campaign-deepfake-ads,reach,"Public reporting identified at least three recent national-Republican synthetic ads by March 2026, but did not provide comparable impression or targeting data.",established_evidence,supported_as_worded,supported_as_worded,measurement,src-us-reuters-midterms-20260328,,Content count is not audience reach.,verified,2026-08-11
clm-us26-011,us-2026-campaign-deepfake-ads,mechanism,"Additional 2026 election ads fabricated audiovisual speech or conduct by Jon Ossoff, John Cornyn, and Ken Paxton, with varying levels of obviousness and disclosure.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-reuters-midterms-20260328,,Record separate items in a later content-level expansion rather than treating them as identical.,verified,2026-08-11
clm-us26-012,us-2026-campaign-deepfake-ads,behavioural_effect,"Among the public sources linked to this record at the cutoff, none measured whether a named 2026 synthetic ad changed candidate evaluations, turnout intention, or vote choice.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,needs_update,2026-08-11
clm-us26-013,us-2026-campaign-deepfake-ads,counterevidence,"Some 2026 campaign-generated videos reproduce genuine prior words, are labelled, or use obvious satire, complicating a binary real-versus-deceptive classification.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-reuters-midterms-20260328,,"Motivates content-level coding for quotation fidelity, disclosure, realism, authorisation, and context.",verified,2026-08-11
clm-us26-014,us-2026-campaign-deepfake-ads,mechanism,"Much of the Talarico attack used genuine older posts, but the synthetic first-person performance added reaction lines and affect not contained in those quoted posts.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-nrsc-talarico-20260311|src-us-reuters-midterms-20260328,,Do not describe every spoken proposition as fabricated; separate authentic underlying text from invented performance and asides.,verified,2026-08-11
clm-us26-015,us-2026-campaign-deepfake-ads,reach,"Among the public sources linked to this record at the cutoff, none provided a preserved, reproducible measure of cross-platform impressions or unique exposure for the NRSC upload.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,The mutable YouTube counter was not retained as evidence without a preserved snapshot. This bounded non-identification statement is not proof that no exposure occurred.,verified,2026-08-11
clm-us26-016,us-2026-campaign-deepfake-ads,control_shift_hypothesis,"Hypothesis: simulated first-person performance could shift control over an opponent's audiovisual representation toward the sponsor; the record observes content production, not voter agency change.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,Analytical hypothesis linked through the pathway to factual claims; it is not direct evidence of an agency outcome.,analytical_hypothesis,2026-08-11
clm-ro-020,ro-2024-presidential-annulment,control_shift_hypothesis,Hypothesis: opaque coordinated promotion and recommender amplification could shift agenda-setting control toward campaign-network operators and ranking systems; the record does not observe individual agency change or establish a causal link to annulment.,researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,"Analytical synthesis of clm-ro-002, clm-ro-003, clm-ro-006, clm-ro-007, and clm-ro-013; not an observed agency outcome.",analytical_hypothesis,2026-08-12
clm-ro-021,ro-2025-presidential-repeat,control_shift_hypothesis,"Hypothesis: synthetic local personas and coordinated distribution could shift control over apparent local consensus toward an unidentified operator; user belief, choice, and electoral effects are not observed.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,Analytical synthesis of clm-ro-014 through clm-ro-018; not an observed agency outcome.,analytical_hypothesis,2026-08-12
clm-md-020,md-2024-presidential-referendum,control_shift_hypothesis,"Hypothesis: covert payments, coordination, and impersonation could shift control over participation or preference toward sponsors; the record supports an inducement attempt but not how recipients voted.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,Analytical synthesis of clm-md-003 through clm-md-011; not an observed agency outcome.,analytical_hypothesis,2026-08-12
clm-md-021,md-2025-parliamentary-linked-network,control_shift_hypothesis,"Hypothesis: centrally tasked paid personas and reciprocal amplification could shift agenda control toward network curators; content and engagement are observed, but user agency and electoral effects are not.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,"Analytical synthesis of clm-md-012, clm-md-015 through clm-md-019; not an observed agency outcome.",analytical_hypothesis,2026-08-12
clm-usca11-001,us-ca11-2026-wiener-conniechan-chatbot,occurrence,Scott Wiener for Congress launched the ConnieChan.ai interactive attack-ad chatbot on 3 August 2026 and removed it on 7 August after public criticism.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-ca11-sfgate-20260804|src-us-ca11-mission-20260807|src-us-ca11-axios-20260811,,"SFGATE described a Monday launch; Mission Local gives 3 August. Some later summaries use 4 August, the first reporting date.",verified,2026-08-12
clm-usca11-002,us-ca11-2026-wiener-conniechan-chatbot,attribution,"The Wiener campaign admitted commissioning, paying for, promoting, and controlling the chatbot; the technical builder was not publicly identified.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-mission-20260807,,"Anthropic supplied a model but is not attributed as campaign designer, approver, or operator.",verified,2026-08-12
clm-usca11-003,us-ca11-2026-wiener-conniechan-chatbot,ai_role,"The campaign said the chatbot used Anthropic's Claude model family to generate interactive responses; the exact version, deployment route, prompt, retrieval system, and guardrails remain unknown.",established_evidence,supported_as_worded,supported_as_worded,assertion_was_made,src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-axios-20260811,,Product family is known; technical configuration is not.,verified,2026-08-12
clm-usca11-004,us-ca11-2026-wiener-conniechan-chatbot,mechanism,The system answered in Connie Chan's first person and dynamically redirected questions toward campaign-selected portrayals of her record.,established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-axios-20260811,,Interactive opponent simulation differs from a static synthetic ad because user questions affect the generated response.,verified,2026-08-12
clm-usca11-005,us-ca11-2026-wiener-conniechan-chatbot,counterevidence,"The site visibly framed itself as AI and parody, said it was not affiliated with or endorsed by Chan, and disclosed that it was paid for by the Wiener campaign.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804,,"Transparent parody weakens a covert-deception theory but does not remove questions about synthetic first-person representation, bias, or data control.",verified,2026-08-12
clm-usca11-006,us-ca11-2026-wiener-conniechan-chatbot,reach,"The campaign reported that the chatbot received 5,058 submitted questions during its first 27 hours.",established_as_campaign_report,supported_as_attributed_measurement,attributed_measurement_only,assertion_was_made,src-us-ca11-mission-20260807|src-us-ca11-axios-20260811,,"Campaign-reported measure. It is not unique users, conversations, voters, district residents, impressions, beliefs, or votes; repeated or adversarial prompts may inflate it.",verified,2026-08-12
clm-usca11-007,us-ca11-2026-wiener-conniechan-chatbot,mechanism,"The privacy policy said conversations could be retained for 90 days and reviewed by a small number of authorised staff to fix bugs, understand queries, and improve campaign content.",established_evidence,supported_as_worded,supported_as_worded,underlying_fact,src-us-ca11-gazetteer-20260804,,"The data warning was in the privacy policy, not prominently repeated in the main chat interface according to the reporting.",verified,2026-08-12
clm-usca11-008,us-ca11-2026-wiener-conniechan-chatbot,response,"On 7 August the campaign removed the chatbot, announced removal of the two billboards, said the experiment had missed the mark, and apologised for causing offence.",established_evidence,supported_as_worded,supported_as_worded,institutional_action,src-us-ca11-mission-20260807|src-us-ca11-axios-20260811,,"Campaign self-termination is a response, not an institutional or legal finding.",verified,2026-08-12
clm-usca11-009,us-ca11-2026-wiener-conniechan-chatbot,behavioural_effect,"Among the public sources linked to this record at the cutoff, none measured whether interaction with the chatbot changed candidate evaluation, trust, issue salience, turnout intention, or vote choice.",not_identified_in_declared_review,not_identified_in_declared_review,review_bounded_absence,review_scope_statement,,,Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.,verified,2026-08-12
clm-usca11-010,us-ca11-2026-wiener-conniechan-chatbot,electoral_effect,"The election had not occurred by the cutoff, and no electoral effect can be inferred from question counts, press attention, criticism, or campaign withdrawal.",open_question,unresolved,none,researcher_inference,,,Prospective guardrail.,verified,2026-08-12
clm-usca11-011,us-ca11-2026-wiener-conniechan-chatbot,control_shift_hypothesis,"Boundary hypothesis: a campaign-controlled opponent chatbot could shift representational and evidentiary control toward its sponsor, but explicit parody, disclosure, unknown system configuration, and unmeasured user effects prevent a finding of agency change.",researcher_hypothesis,hypothesis_not_tested,not_tested,researcher_inference,,,Analytical hypothesis linked through the pathway to factual claims; it is not direct evidence of an agency outcome.,analytical_hypothesis,2026-08-12