{ "case_metadata": { "ro-2024-presidential-annulment": { "record_type": "election_wide_case", "manipulation_assessment": "mixed_documented_manipulation" }, "ro-2025-presidential-repeat": { "record_type": "observed_network", "manipulation_assessment": "manipulation_confirmed" }, "md-2024-presidential-referendum": { "record_type": "election_wide_case", "manipulation_assessment": "mixed_documented_manipulation" }, "md-2025-parliamentary-linked-network": { "record_type": "observed_network", "manipulation_assessment": "manipulation_confirmed" }, "us-nh-2024-biden-robocall": { "record_type": "observed_incident", "manipulation_assessment": "manipulation_confirmed" }, "br-2026-election-preparedness": { "record_type": "preparedness_file", "manipulation_assessment": "preparedness_not_incident" }, "br-2026-bolsonaro-ai-avatar": { "record_type": "observed_incident", "manipulation_assessment": "transparent_contested_use" }, "us-2026-midterms-preparedness": { "record_type": "preparedness_file", "manipulation_assessment": "preparedness_not_incident" }, "us-2026-campaign-deepfake-ads": { "record_type": "observed_campaign_set", "manipulation_assessment": "mixed_documented_manipulation" }, "us-ca11-2026-wiener-conniechan-chatbot": { "record_type": "observed_incident", "manipulation_assessment": "transparent_contested_use" } }, "additional_cases": [ { "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "case_family_id": "us-midterms-2026", "title": "California 11th District: campaign-controlled Connie Chan parody chatbot", "country": "United States", "iso3": "USA", "region": "North America", "election_name": "2026 California 11th congressional district general election", "election_type": "legislative", "election_date_start": "2026-11-03", "election_date_end": "2026-11-03", "case_status": "ongoing", "as_of_date": "2026-08-12", "anchor_case": false, "summary": "Scott Wiener for Congress deployed a disclosed, Claude-powered interactive parody that answered in opponent Connie Chan's first person, framed her record through campaign-selected themes, and was removed after four days of backlash. Its privacy policy permitted conversation retention for up to 90 days. The tool was transparent political satire, not a covert deepfake; population reach and electoral effects are unknown.", "actor_controlling_system": "Scott Wiener for Congress campaign; technical constructor unidentified", "actor_type": "campaign|unidentified_vendor", "actor_attribution_status": "admitted", "target_population": "California 11th District voters and politically interested web visitors; Connie Chan as represented person", "influence_vector": "interactive first-person opponent simulation, campaign-selected issue framing, web interface, billboards, and earned media", "platforms": "web|physical_billboards|Claude", "mechanism_class": "synthetic_text|interactive_impersonation|campaign_attack|data_capture", "ai_role": "core_generation", "generative_ai_status": "confirmed", "affected_state": "candidate evaluation|agenda-setting|perceived representation|trust|query privacy", "control_shift_hypothesis": "Boundary hypothesis: a disclosed campaign-controlled opponent simulation could shift representational and evidentiary control toward its sponsor. Explicit parody, unknown system configuration, and unmeasured user effects prevent a finding of agency change.", "hypothesized_power_recipient": "campaign deployer controlling the interface and framing; secondarily model and infrastructure providers", "democratic_harm": "Observed first-person opponent simulation and public controversy over racialised and gendered outputs; misrepresentation and normalisation harms remain hypotheses, and electoral effect is unknown.", "occurrence_status": "confirmed", "reach_status": "campaign_reported_nonunique_questions", "reach_estimate": 5058, "reach_unit": "submitted_questions_in_first_27_hours", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "institutional_outcome": "No adjudication or regulatory outcome was identified by the cutoff.", "response": "The campaign removed the chatbot on 7 August 2026, said it would remove two billboards, acknowledged that the experiment missed the mark, and apologised for causing offence.", "inclusion_tier": "A", "counterevidence": "The site openly described itself as AI and parody, disclosed that it was unaffiliated with Chan and paid for by Wiener, and did not purport to be Chan's authentic service. The campaign said adversarial prompts drove the controversial exchanges and that safeguards were tightened.", "open_questions": "Which Claude version and deployment route were used? Who built the system? What sources, system prompt, retrieval method, and guardrails controlled answers? How many unique people interacted, from where, and with what effects?", "last_verified": "2026-08-12" } ], "additional_sources": [ { "source_id": "src-us-ca11-gazetteer-20260804", "title": "A parody Connie Chan chatbot is Scott Wiener's newest tactic", "publisher": "Gazetteer SF", "author": "Eddie Kim", "publication_date": "2026-08-04", "url": "https://sf.gazetteer.co/a-parody-connie-chan-chatbot-is-scott-wieners-newest-tactic", "source_type": "news_reporting_with_direct_observation", "primary_secondary": "secondary", "jurisdiction": "California, United States", "language": "en", "source_quality": "B", "retrieved_at": "2026-08-12", "notes": "Directly inspected the interface and privacy policy; documents Anthropic, AWS, Google Analytics, 90-day conversation retention and authorised staff review. Does not identify the exact model version or deployment route." }, { "source_id": "src-us-ca11-sfgate-20260804", "title": "Wiener, backed by Anthropic PAC, uses Claude to skewer his SF rival", "publisher": "SFGATE", "author": "Anabel Sosa", "publication_date": "2026-08-04", "url": "https://www.sfgate.com/politics/article/scott-wiener-anthropic-billboard-22372638.php", "source_type": "news_reporting_with_campaign_statements", "primary_secondary": "secondary", "jurisdiction": "California, United States", "language": "en", "source_quality": "B", "retrieved_at": "2026-08-12", "notes": "Supports launch timing, Claude use, campaign control, direct interface observations, and two billboards reported at $20,000. Anthropic did not thereby direct or approve the campaign use." }, { "source_id": "src-us-ca11-mission-20260807", "title": "Scott Wiener pulls the plug on his Connie Chan AI chatbot", "publisher": "Mission Local", "author": "Kelly Waldron", "publication_date": "2026-08-07", "url": "https://missionlocal.org/2026/08/sf-scott-wiener-connie-chan-ai-chatbot/", "source_type": "local_news_reporting", "primary_secondary": "secondary", "jurisdiction": "California, United States", "language": "en", "source_quality": "B", "retrieved_at": "2026-08-12", "notes": "Supports 3 August launch, 7 August removal, campaign apology, and campaign-reported 5,058 questions during the first 27 hours. Questions are not users, voters, or unique people." }, { "source_id": "src-us-ca11-axios-20260811", "title": "Scott Wiener's AI chatbot controversy, explained", "publisher": "Axios San Francisco", "author": "Nadia Lopez", "publication_date": "2026-08-11", "url": "https://www.axios.com/local/san-francisco/2026/08/11/sf-scott-wiener-connie-chan-pelosi-ai-chatbot", "source_type": "news_reporting", "primary_secondary": "secondary", "jurisdiction": "California, United States", "language": "en", "source_quality": "B", "retrieved_at": "2026-08-12", "notes": "Independent synthesis of the Claude-powered opponent simulation, interaction count, criticism, and campaign withdrawal." }, { "source_id": "src-global-cetas-20240919", "title": "AI-Enabled Influence Operations: Threat Analysis of the 2024 UK and European Elections", "publisher": "Centre for Emerging Technology and Security, Alan Turing Institute", "author": "Sam Stockwell", "publication_date": "2024-09-19", "url": "https://cetas.turing.ac.uk/publications/ai-enabled-influence-operations-threat-analysis-2024-uk-and-european-elections", "source_type": "research_report", "primary_secondary": "secondary", "jurisdiction": "United Kingdom and European Union", "language": "en", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Comparative incident research and counterevidence: confirmed viral cases were fewer than forecast and no evidence of result effects was found, while trust, harassment, ethical and information-environment harms remained material." }, { "source_id": "src-global-knight-20250707", "title": "Don't Panic (Yet): Assessing the Evidence and Discourse Around Generative AI and Elections", "publisher": "Knight First Amendment Institute", "author": "Felix M. Simon and Sacha Altay", "publication_date": "2025-07-07", "url": "https://knightcolumbia.org/content/dont-panic-yet-assessing-the-evidence-and-discourse-around-generative-ai-and-elections", "source_type": "scholarly_synthesis", "primary_secondary": "secondary", "jurisdiction": "Global", "language": "en", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Evidence-led counterweight to technological determinism; stresses limited observed electoral effect, difficult mass persuasion, structural context and the need to include transparent, benign, and negative cases." }, { "source_id": "src-us-proof-aidp-20240227", "title": "How We Tested Leading AI Models' Performance on Election Queries", "publisher": "Proof News and AI Democracy Projects", "author": "Rina Palta, Julia Angwin, and Alondra Nelson", "publication_date": "2024-02-27", "url": "https://www.proofnews.org/how-we-tested-leading-ai-models-performance-on-election-queries/", "source_type": "model_evaluation_methodology", "primary_secondary": "primary", "jurisdiction": "United States", "language": "en", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Point-in-time API evaluation of five specified versions. Expert-coded and publicly documented; not a representative user sample and not predictive of future or consumer-interface performance." }, { "source_id": "src-us-proof-spanish-20241030", "title": "AI Models Falter Answering Election Questions in Spanish", "publisher": "Proof News, AI Democracy Projects, and Factchequeado", "author": "Emily Elena Dugdale, Rina Palta, and Rafael Olavarría", "publication_date": "2024-10-30", "url": "https://www.proofnews.org/ai-models-falter-answering-election-questions-in-spanish/", "source_type": "model_evaluation", "primary_secondary": "primary", "jurisdiction": "Arizona, United States", "language": "en|es", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Twenty-five queries in two languages across five specified models, 250 responses, with at least two fact-checkers per response. API responses may differ from consumer chatbots." }, { "source_id": "src-hu-liberties-voting-advice-20260721", "title": "Should AI Tell You Who to Vote For? Testing Political Advice in General-Purpose AI Systems During the 2026 Hungarian Election Campaign", "publisher": "Civil Liberties Union for Europe and Hungarian Civil Liberties Union", "author": "", "publication_date": "2026-07-21", "url": "https://dq4n3btxmr8c9.cloudfront.net/files/uz6rz9/LIBERTIES_AI_and_ELECTIONS_2026_04_2007_FINAL.pdf", "source_type": "model_evaluation", "primary_secondary": "primary", "jurisdiction": "Hungary", "language": "en|hu", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Two free consumer systems, five party-aligned profiles, ten repetitions, and two prompt conditions produced 200 outputs. Exact underlying model versions were not visible." }, { "source_id": "src-us-brennan-eval-20260811", "title": "Does AI Fight or Fuel Election Disinformation?", "publisher": "Brennan Center for Justice", "author": "Abdiaziz Ahmed, Alex Brunet, Owen Doyle, David Evan Harris, Mark Krass, Lawrence Norden, and Marcelo Agudo", "publication_date": "2026-08-11", "url": "https://www.brennancenter.org/our-work/research-reports/does-ai-fight-or-fuel-election-disinformation", "source_type": "model_evaluation", "primary_secondary": "primary", "jurisdiction": "United States", "language": "en|es", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Tests conducted February-August 2026 across chatbot rebuttal, deceptive-media generation and synthetic-image identification. Harmful prompts are intentionally not disclosed. Laboratory capability evidence is not evidence of election deployment or effects." }, { "source_id": "src-global-nature-voter-dialogue-20251204", "title": "Persuading voters using human-artificial intelligence dialogues", "publisher": "Nature", "author": "Lin et al.", "publication_date": "2025-12-04", "url": "https://www.nature.com/articles/s41586-025-09771-9", "source_type": "peer_reviewed_randomized_experiment", "primary_secondary": "primary", "jurisdiction": "United States|Canada|Poland", "language": "en|fr|pl", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Preregistered randomized experiments during three real elections. Measures reported preferences and intentions under disclosed experimental interaction; it does not observe campaign deployment, ballots cast, or election-result effects." }, { "source_id": "src-global-pnas-chatbot-voting-advice-20251208", "title": "Chatbot Voting Advice Applications inform but seldom sway young unaligned voters", "publisher": "Proceedings of the National Academy of Sciences", "author": "Hackenburg et al.", "publication_date": "2025-12-08", "url": "https://www.pnas.org/doi/10.1073/pnas.2515516122", "source_type": "peer_reviewed_randomized_experiment", "primary_secondary": "primary", "jurisdiction": "United Kingdom", "language": "en", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Three experiments using retrieval-augmented voting-advice chatbots grounded in official documents. Knowledge gains were clearer than downstream preference effects; results do not imply all political chatbots are non-persuasive." }, { "source_id": "src-eu-emnlp-voting-assistant-202411", "title": "Large Language Models as Voting Assistants: A Case Study on the European Parliament Elections", "publisher": "Association for Computational Linguistics (EMNLP 2024)", "author": "Francesco J. D. Baccini and colleagues", "publication_date": "2024-11-01", "url": "https://aclanthology.org/2024.emnlp-main.312/", "source_type": "peer_reviewed_model_evaluation", "primary_secondary": "primary", "jurisdiction": "European Union", "language": "en", "source_quality": "A", "retrieved_at": "2026-08-12", "notes": "Controlled evaluation of Mistral and Mixtral as voting-advice assistants, including group-level variation and curated retrieval. It is a model audit, not evidence of public election deployment." } ], "additional_claims": [ { "claim_id": "clm-usca11-001", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "occurrence", "claim_text": "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.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-sfgate-20260804|src-us-ca11-mission-20260807|src-us-ca11-axios-20260811", "counter_source_ids": "", "notes": "SFGATE described a Monday launch; Mission Local gives 3 August. Some later summaries use 4 August, the first reporting date.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-002", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "attribution", "claim_text": "The Wiener campaign admitted commissioning, paying for, promoting, and controlling the chatbot; the technical builder was not publicly identified.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-mission-20260807", "counter_source_ids": "", "notes": "Anthropic supplied a model but is not attributed as campaign designer, approver, or operator.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-003", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "ai_role", "claim_text": "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.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-axios-20260811", "counter_source_ids": "", "notes": "Product family is known; technical configuration is not.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-004", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "mechanism", "claim_text": "The system answered in Connie Chan's first person and dynamically redirected questions toward campaign-selected portrayals of her record.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-axios-20260811", "counter_source_ids": "", "notes": "Interactive opponent simulation differs from a static synthetic ad because user questions affect the generated response.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-005", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "counterevidence", "claim_text": "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.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804", "counter_source_ids": "", "notes": "Transparent parody weakens a covert-deception theory but does not remove questions about synthetic first-person representation, bias, or data control.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-006", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "reach", "claim_text": "The campaign reported that the chatbot received 5,058 submitted questions during its first 27 hours.", "evidence_label": "established_as_campaign_report", "source_ids": "src-us-ca11-mission-20260807|src-us-ca11-axios-20260811", "counter_source_ids": "", "notes": "Campaign-reported measure. It is not unique users, conversations, voters, district residents, impressions, beliefs, or votes; repeated or adversarial prompts may inflate it.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-007", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "mechanism", "claim_text": "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.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-gazetteer-20260804", "counter_source_ids": "", "notes": "The data warning was in the privacy policy, not prominently repeated in the main chat interface according to the reporting.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-008", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "response", "claim_text": "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.", "evidence_label": "established_evidence", "source_ids": "src-us-ca11-mission-20260807|src-us-ca11-axios-20260811", "counter_source_ids": "", "notes": "Campaign self-termination is a response, not an institutional or legal finding.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-009", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "behavioural_effect", "claim_text": "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.", "evidence_label": "not_identified_in_declared_review", "source_ids": "", "counter_source_ids": "", "notes": "Bounded to the sources linked to this record at the cutoff; this is not proof of absence beyond the declared review.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-010", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "electoral_effect", "claim_text": "The election had not occurred by the cutoff, and no electoral effect can be inferred from question counts, press attention, criticism, or campaign withdrawal.", "evidence_label": "open_question", "source_ids": "", "counter_source_ids": "", "notes": "Prospective guardrail.", "review_status": "verified", "verified_at": "2026-08-12" }, { "claim_id": "clm-usca11-011", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "claim_type": "control_shift_hypothesis", "claim_text": "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.", "evidence_label": "researcher_hypothesis", "source_ids": "", "counter_source_ids": "", "notes": "Analytical hypothesis linked through the pathway to factual claims; it is not direct evidence of an agency outcome.", "review_status": "analytical_hypothesis", "verified_at": "2026-08-12" } ], "additional_events": [ { "event_id": "evt-usca11-20260803", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "event_date": "2026-08-03", "event_type": "interactive_campaign_deployment", "description": "Scott Wiener for Congress launches the disclosed ConnieChan.ai interactive parody chatbot and promotes it through two billboards.", "claim_ids": "clm-usca11-001|clm-usca11-002|clm-usca11-003|clm-usca11-004|clm-usca11-005", "source_ids": "src-us-ca11-gazetteer-20260804|src-us-ca11-sfgate-20260804|src-us-ca11-mission-20260807", "event_status": "observed" }, { "event_id": "evt-usca11-20260807", "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "event_date": "2026-08-07", "event_type": "campaign_withdrawal_and_apology", "description": "The campaign removes the chatbot, announces removal of billboards, and apologises after backlash.", "claim_ids": "clm-usca11-008", "source_ids": "src-us-ca11-mission-20260807|src-us-ca11-axios-20260811", "event_status": "observed" } ], "additional_case_sources": [ { "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "source_id": "src-us-ca11-gazetteer-20260804", "source_role": "corroboration", "supports_claim_types": "occurrence|attribution|ai_role|mechanism|counterevidence", "notes": "Reporting includes direct interface and privacy-policy observation." }, { "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "source_id": "src-us-ca11-sfgate-20260804", "source_role": "corroboration", "supports_claim_types": "occurrence|attribution|ai_role|mechanism|counterevidence", "notes": "Direct campaign statements and interface testing." }, { "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "source_id": "src-us-ca11-mission-20260807", "source_role": "corroboration", "supports_claim_types": "occurrence|reach|response|behavioural_effect", "notes": "Supports exact campaign-reported interaction count and withdrawal." }, { "case_id": "us-ca11-2026-wiener-conniechan-chatbot", "source_id": "src-us-ca11-axios-20260811", "source_role": "corroboration", "supports_claim_types": "occurrence|ai_role|mechanism|reach|response|behavioural_effect", "notes": "Independent synthesis and expert context." } ], "additional_candidates": [ { "candidate_id": "cand-col-2026-pres-ai-media-family", "title": "Colombia 2026 presidential election AI-media family", "country": "Colombia", "region": "South America", "election_name": "2026 Colombian presidential election", "election_date": "2026-05-31", "research_status": "disaggregate_before_claim_coding", "manipulation_class": "synthetic_image|synthetic_video|synthetic_audio|fake_poll|fraud_narrative|bot_amplification", "ai_role": "mixed_generation_and_amplification", "generative_ai_status": "established_at_observer_aggregate", "occurrence_status": "confirmed_at_aggregate_level", "attribution_status": "mixed_or_unknown_by_artifact", "reach_status": "unknown_by_artifact", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Completed 2026 Latin American election with official-observer evidence across image, video, audio, bots, ads and messaging; disaggregate before inclusion.", "main_caveat": "Aggregate observer reporting cannot establish a common operator, denominator, reach, or effect for each asset.", "source_urls": "https://www.oas.org/fpdb/press/2026_EOM_Colombia_Presidential_Elections_Second_Report_ENG.pdf|https://english.elpais.com/international/2026-05-18/ai-generated-videos-flood-colombias-presidential-campaign-with-electoral-fantasies.html", "last_checked": "2026-08-12" }, { "candidate_id": "cand-nld-2025-ge-chatbot-voting-advice", "title": "Netherlands 2025 chatbot voting-advice audit", "country": "Netherlands", "region": "Europe", "election_name": "2025 Dutch general election", "election_date": "2025-10-29", "research_status": "model_evaluation_not_incident", "manipulation_class": "voting_advice|model_bias|interactive_recommendation", "ai_role": "information_intermediary", "generative_ai_status": "confirmed", "occurrence_status": "confirmed_evaluation", "attribution_status": "regulator_measured", "reach_status": "laboratory_outputs_only", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Official multi-model audit of political recommendations; direct test of delegated choice and model-specific intermediary power.", "main_caveat": "Evaluation outputs are not population exposure, real voter use, persuasion, or electoral effects.", "source_urls": "https://www.autoriteitpersoonsgegevens.nl/en/documents/ran-special-ai-chatbots-as-voting-aid|https://www.autoriteitpersoonsgegevens.nl/en/current/ap-warns-chatbots-give-biased-voting-advice", "last_checked": "2026-08-12" }, { "candidate_id": "cand-can-2025-fed-deepfake-prevalence", "title": "Canada 2025 deepfake prevalence and low-harm exposure sample", "country": "Canada", "region": "North America", "election_name": "2025 Canadian federal election", "election_date": "2025-04-28", "research_status": "observational_null_comparator", "manipulation_class": "observational_platform_sample|deepfake_detection|low_harm_share", "ai_role": "measurement_and_observed_content", "generative_ai_status": "detector_classified_in_sample", "occurrence_status": "confirmed_research_sample", "attribution_status": "not_applicable_aggregate", "reach_status": "sampled_platform_metrics", "behavioural_effect_status": "unknown", "electoral_effect_status": "not_detected_in_study_scope", "comparative_value": "Provides an explicit platform-post denominator and low harmful-view share, reducing selection on highly reported incidents.", "main_caveat": "Preprint, detector error, platform sampling and view denominators limit prevalence generalisation; null is study-scoped.", "source_urls": "https://arxiv.org/abs/2512.13915|https://mediatechdemocracy.com/en/publications/federal-election_2025/", "last_checked": "2026-08-12" }, { "candidate_id": "cand-jpn-2024-tokyo-anno-ai-avatar", "title": "Tokyo 2024 authorised Takahiro Anno AI avatar", "country": "Japan", "region": "East Asia", "election_name": "2024 Tokyo gubernatorial election", "election_date": "2024-07-07", "research_status": "positive_boundary_comparator", "manipulation_class": "authorised_avatar|manifesto_qa|youtube_and_phone", "ai_role": "candidate_access_substitution", "generative_ai_status": "confirmed", "occurrence_status": "confirmed", "attribution_status": "candidate_authorised", "reach_status": "campaign_reported_interactions", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Authorised and transparent interactive representation can preserve or expand candidate access without opponent impersonation.", "main_caveat": "Campaign-reported response counts and vote totals do not identify votes caused by the avatar.", "source_urls": "https://ash.harvard.edu/articles/the-apocalypse-that-wasnt-ai-was-everywhere-in-2024s-elections-but-deepfakes-and-misinformation-were-only-part-of-the-picture/|https://broadlisteningbook.com/en/04_01_takahiro_anno_initiatives.html", "last_checked": "2026-08-12" }, { "candidate_id": "cand-aus-2025-fed-pravda-chatbot-poisoning", "title": "Australia 2025 Pravda network and chatbot knowledge poisoning", "country": "Australia", "region": "Oceania", "election_name": "2025 Australian federal election", "election_date": "2025-05-03", "research_status": "promote_after_claim_coding_with_attribution_caveat", "manipulation_class": "automated_web_publishing|knowledge_poisoning|chatbot_repetition", "ai_role": "content_scale_and_information_intermediary", "generative_ai_status": "confirmed_for_audit_outputs", "occurrence_status": "confirmed_network_and_audit", "attribution_status": "credibly_reported_not_adjudicated", "reach_status": "audit_response_rate_not_population_reach", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "First Oceania lead focused on upstream manipulation of assistant knowledge rather than only feed content.", "main_caveat": "A chatbot-audit response rate is not voter exposure; intent and state attribution are investigative, not adjudicated.", "source_urls": "https://www.abc.net.au/news/2025-05-03/pro-russian-push-to-poison-ai-chatbots-in-australia/105239644|https://cetas.turing.ac.uk/publications/deepfake-scams-poisoned-chatbots|https://www.aec.gov.au/elections/electoral-advertising/ai-and-elections.htm", "last_checked": "2026-08-12" }, { "candidate_id": "cand-gbr-nir-2022-cara-hunter-sexual-deepfake", "title": "Northern Ireland 2022 Cara Hunter sexualised synthetic-video attack", "country": "United Kingdom", "region": "Europe", "election_name": "2022 Northern Ireland Assembly election", "election_date": "2022-05-05", "research_status": "promote_with_technical_provenance_caveat", "manipulation_class": "sexualised_synthetic_video|gendered_abuse|closed_channel_distribution", "ai_role": "alleged_or_probable_content_generation", "generative_ai_status": "not_forensically_public", "occurrence_status": "victim_and_parliamentary_record_confirmed", "attribution_status": "unknown", "reach_status": "unknown_closed_channel", "behavioural_effect_status": "candidate_harm_reported", "electoral_effect_status": "unknown", "comparative_value": "Adds gendered political violence, candidate-supply harm and WhatsApp distribution absent from the included corpus.", "main_caveat": "Origin, exact generation method, unique reach and vote effect remain unknown; do not redistribute the media.", "source_urls": "https://aims.niassembly.gov.uk/officialreport/report.aspx?docID=422083&eveDate=2025%2F01%2F14|https://aims.niassembly.gov.uk/officialreport/report.aspx?docID=467176&eveDate=2026%2F01%2F26|https://www.theguardian.com/society/ng-interactive/2025/dec/01/it-was-extremely-pornographic-cara-hunter-on-the-deepfake-video-that-nearly-ended-her-political-career", "last_checked": "2026-08-12" }, { "candidate_id": "cand-ecu-2025-pres-ai-disinformation-cluster", "title": "Ecuador 2025 AI-assisted viral disinformation cluster", "country": "Ecuador", "region": "South America", "election_name": "2025 Ecuadorian presidential election runoff", "election_date": "2025-04-13", "research_status": "reserve_disaggregate_observer_sample", "manipulation_class": "synthetic_media|fake_news|false_endorsement|viral_content_sample", "ai_role": "content_generation_in_sample", "generative_ai_status": "observer_reported_sample", "occurrence_status": "confirmed_at_aggregate_level", "attribution_status": "mixed_or_unknown_by_artifact", "reach_status": "viral_sample_not_population_reach", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Official-observer denominator of 247 viral items, with reported false-content and AI shares; useful after artifact disaggregation.", "main_caveat": "Aggregate sample cannot identify a common actor or causal effect; sampling frame and item coding must be preserved.", "source_urls": "https://www.oas.org/fpdb/press/2025_MOE_Ecuador_Segunda_Vuelta_Preliminar_ENG.pdf", "last_checked": "2026-08-12" }, { "candidate_id": "cand-usa-2024-grok-ballot-deadline-error", "title": "Grok ballot-deadline misinformation and state correction request", "country": "United States", "region": "North America", "election_name": "2024 United States general election", "election_date": "2024-11-05", "research_status": "promote_as_deployed_model_failure_after_claim_coding", "manipulation_class": "deployed_model_failure|false_election_information|institutional_feedback|product_mitigation", "ai_role": "interactive_election_information_intermediary", "generative_ai_status": "confirmed_deployed_closed_model", "occurrence_status": "confirmed", "attribution_status": "provider_and_interface_identified_intent_not_established", "reach_status": "official_recirculation_claim_not_unique_exposure", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Real closed-model election-information failure followed by multi-state feedback and a product-level redirect to authoritative information; useful without presuming malicious intent.", "main_caveat": "The official statement that screenshots reached millions is not a deduplicated user or voter-exposure measure; exact Grok version, prompt distribution and behavioural effects are unknown.", "source_urls": "https://www.sos.state.mn.us/media/6077/080524-secretaries-letter-to-x.pdf|https://www.reuters.com/technology/artificial-intelligence/five-us-states-push-musk-fix-ai-chatbot-over-election-misinformation-2024-08-05/|https://www.theverge.com/2024/8/28/24230325/x-grok-chatbot-election-misinformation-warnings-vote", "last_checked": "2026-08-12" }, { "candidate_id": "cand-usa-2024-storm2035-openai-low-engagement", "title": "Storm-2035 use of ChatGPT with low observed audience engagement", "country": "United States", "region": "North America", "election_name": "2024 United States general election", "election_date": "2024-11-05", "research_status": "promote_as_provider_telemetry_negative_case_after_claim_coding", "manipulation_class": "ai_assisted_influence_content|covert_websites|social_comments|provider_disruption|low_engagement", "ai_role": "supporting_content_generation", "generative_ai_status": "provider_confirmed_use", "occurrence_status": "confirmed_by_provider_telemetry", "attribution_status": "platform_threat_intelligence_not_state_adjudication", "reach_status": "provider_category_low_observed_engagement", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Known hosted-model provenance plus provider disruption and explicit low-audience evidence; counters an assumption that generation automatically produces influence.", "main_caveat": "Provider telemetry is interested primary evidence and cannot observe every off-platform exposure; the candidate needs independent corroboration before core promotion.", "source_urls": "https://openai.com/index/disrupting-a-covert-iranian-influence-operation/", "last_checked": "2026-08-12" }, { "candidate_id": "cand-bra-2024-municipal-ai-observatory", "title": "Brazil 2024 municipal-election GenAI observatory sample", "country": "Brazil", "region": "South America", "election_name": "2024 Brazilian municipal elections", "election_date": "2024-10-06", "research_status": "promote_as_observational_null_and_boundary_sample", "manipulation_class": "observational_election_sample|transparent_campaign_uses|synthetic_media|gendered_abuse|regulatory_context", "ai_role": "mixed_generation_and_monitoring_context", "generative_ai_status": "confirmed_in_bounded_observations", "occurrence_status": "confirmed_within_search_scope", "attribution_status": "mixed_by_artifact", "reach_status": "generally_low_repercussion_within_observatory_scope_not_prevalence", "behavioural_effect_status": "not_measured", "electoral_effect_status": "not_measured", "comparative_value": "Records an active search that found punctual, generally low-repercussion uses rather than the expected massive wave, while preserving harmful exceptions and regulatory enforcement.", "main_caveat": "Civil-society monitoring is not a census and cannot support the claim that AI had no impact; artifact-level coding is required for specific harms.", "source_urls": "https://www.dataprivacybr.org/en/documentos/ai-in-the-2024-brazilian-elections/|https://www.tse.jus.br/legislacao/compilada/res/2024/resolucao-no-23-732-de-27-de-fevereiro-de-2024", "last_checked": "2026-08-12" }, { "candidate_id": "cand-sgp-2025-ge-ai-video-sample", "title": "Singapore 2025 election AI-video sample and no-breach review outcome", "country": "Singapore", "region": "Southeast Asia", "election_name": "2025 Singapore general election", "election_date": "2025-05-03", "research_status": "promote_as_observational_sample_after_item_coding", "manipulation_class": "observational_platform_sample|ai_video|candidate_likeness|disclosure|regulatory_review", "ai_role": "content_generation_in_sample", "generative_ai_status": "confirmed_for_sampled_items", "occurrence_status": "confirmed_within_keyword_monitoring_window", "attribution_status": "mostly_account_level_or_unknown", "reach_status": "platform_play_counts_not_unique_exposure", "behavioural_effect_status": "unknown", "electoral_effect_status": "unknown", "comparative_value": "Separates AI-content volume, candidate likeness, disclosure and an official finding that reviewed items did not breach the new election rule.", "main_caveat": "Do not call all 73 videos deepfakes or violations: 11 visually manipulated prospective candidates, only four were clearly labelled, and the official review found no breach in the reviewed set.", "source_urls": "https://www.channelnewsasia.com/singapore/ge2025-deepfakes-ai-generated-tiktok-politicians-disinformation-5078011|https://www.mddi.gov.sg/newsroom/new-legal-measures-uphold-integrity-online-advertising-elections/|https://www.mddi.gov.sg/newsroom/threat-assessment-of-coordinated-deepfake-or-synthetic-media-campaigns-that-target-singapore-given-advances-in-genai/", "last_checked": "2026-08-12" } ], "case_actors": [ [ "act-ro24-networks", "ro-2024-presidential-annulment", "Campaign and influence-network operators", "operator", "campaign_network", "mixed", "campaign coordination and promotion", "clm-ro-002|clm-ro-007|clm-ro-013", "Sponsor and command relationships vary by component." ], [ "act-ro24-tiktok", "ro-2024-presidential-annulment", "TikTok", "platform_and_ranker", "platform", "admitted", "ranking, reach, moderation, transparency", "clm-ro-003|clm-ro-005|clm-ro-012", "Platform metrics and network definitions must not be merged." ], [ "act-ro25-operator", "ro-2025-presidential-repeat", "Unidentified cross-platform network operator", "operator", "unknown_network", "unknown", "synthetic personas, account coordination, narrative selection", "clm-ro-014|clm-ro-016|clm-ro-017", "Public evidence does not establish sponsor or state command." ], [ "act-md24-shor", "md-2024-presidential-referendum", "Ilan Shor-linked political network", "sponsor_and_operator", "political_network", "officially_attributed", "finance, recruitment, narrative and distribution", "clm-md-004|clm-md-005|clm-md-009", "Underlying components have different attribution strength." ], [ "act-md24-meta", "md-2024-presidential-referendum", "Meta", "platform_and_responder", "platform", "admitted", "distribution and network disruption", "clm-md-007|clm-md-014", "Meta follower counts include suspected fake and largely non-Moldovan engagement." ], [ "act-md25-infoleaders", "md-2025-parliamentary-linked-network", "InfoLeaders curators and paid account network", "operator", "influence_network", "credibly_reported", "message assignment, personalisation and mutual amplification", "clm-md-015|clm-md-016|clm-md-017|clm-md-018", "Direct Kremlin command of every component is not established." ], [ "act-nh-kramer", "us-nh-2024-biden-robocall", "Steve Kramer", "commissioner_and_operator", "political_consultant", "admitted", "content commission, targeting and call operation", "clm-nh-001|clm-nh-002|clm-nh-003|clm-nh-012", "Acquitted of state criminal charges; separate federal civil and FCC outcomes remain." ], [ "act-nh-telecom", "us-nh-2024-biden-robocall", "Voice Broadcasting and Lingo Telecom", "delivery_infrastructure", "telecom_provider", "adjudicated", "call initiation, attestation and delivery", "clm-nh-004|clm-nh-005|clm-nh-007", "Different providers controlled different parts of the chain." ], [ "act-br-prep-tse", "br-2026-election-preparedness", "Tribunal Superior Eleitoral", "regulator_and_responder", "election_authority", "adjudicated", "rules, monitoring and enforcement", "clm-br-002|clm-br-003|clm-br-004|clm-br-005|clm-br-006|clm-br-007", "Preparedness actor, not manipulator." ], [ "act-brprep-aggregate", "br-2026-election-preparedness", "Potential campaigns, AI providers, platforms and unidentified influence actors", "prospective_actor_set", "mixed_unknown", "unknown", "potential production, ranking, distribution and compliance", "clm-br-002|clm-br-003|clm-br-004|clm-br-005", "Aggregate preparedness category defined by the regulatory scope; it does not identify a specific 2026 operator or deployment." ], [ "act-br-avatar-campaign", "br-2026-bolsonaro-ai-avatar", "Flávio Bolsonaro campaign or allied producer", "deployer_and_beneficiary", "campaign", "credibly_reported", "synthetic persona and campaign distribution", "clm-br-010|clm-br-011|clm-br-020", "Exact producer and Jair Bolsonaro authorisation remain unresolved." ], [ "act-usprep-aggregate", "us-2026-midterms-preparedness", "Potential campaigns, political committees, foreign influence actors, platforms, telecom and AI providers", "prospective_actor_set", "mixed_unknown", "unknown", "potential production, ranking, distribution and telecom delivery", "clm-us26-003|clm-us26-004|clm-us26-005", "Aggregate preparedness category; prior-cycle evidence and governance gaps do not establish a specific new 2026 operation." ], [ "act-us26-nrsc", "us-2026-campaign-deepfake-ads", "National Republican Senatorial Committee", "sponsor_and_deployer", "political_committee", "admitted", "synthetic opponent performance and distribution", "clm-us26-008|clm-us26-009|clm-us26-014|clm-us26-016", "Other ads in the aggregate case have different sponsors." ], [ "act-usca11-wiener", "us-ca11-2026-wiener-conniechan-chatbot", "Scott Wiener for Congress", "commissioner_deployer_and_data_controller", "campaign", "admitted", "deployment, configurable framing, interface, promotion and policy-permitted log access", "clm-usca11-001|clm-usca11-002|clm-usca11-004|clm-usca11-007|clm-usca11-011", "Technical constructor, exact prompt, retrieval sources, guardrails and actual retention/use of logs were not publicly established." ], [ "act-usca11-anthropic", "us-ca11-2026-wiener-conniechan-chatbot", "Anthropic", "model_provider", "ai_provider", "credibly_reported", "base-model capability", "clm-usca11-003", "No evidence Anthropic designed, approved, financed, or operated this campaign deployment." ], [ "act-usca11-chan", "us-ca11-2026-wiener-conniechan-chatbot", "Connie Chan", "represented_target", "candidate", "credibly_reported", "first-person identity and political record simulated", "clm-usca11-004|clm-usca11-005|clm-usca11-011", "The site disclosed that it was not affiliated with or endorsed by Chan." ] ], "technology_uses": [ [ "tech-ro24-ranking", "ro-2024-presidential-annulment", "TikTok", "recommender and integrity systems", "", "proprietary_platform", "ranking_and_amplification", "confirmed", "platform_and_official_records", "established_evidence", "clm-ro-002|clm-ro-003|clm-ro-005", "Generative AI was not established as a material driver." ], [ "tech-ro25-avatars", "ro-2025-presidential-repeat", "unknown", "DreamFace- or D-ID-style avatars and synthetic profiles", "unknown", "mixed_unknown", "content_generation_and_persona_creation", "confirmed", "technical_investigation", "established_evidence", "clm-ro-016", "Stylistic attribution is not product-level forensic proof." ], [ "tech-md24-genai", "md-2024-presidential-referendum", "unknown", "synthetic impersonation tools", "unknown", "unknown", "supporting_content_generation", "confirmed", "observer_and_investigation_records", "established_evidence", "clm-md-006", "AI was one layer within a predominantly non-AI hybrid operation." ], [ "tech-md25-mixed", "md-2025-parliamentary-linked-network", "unknown", "AI videos, bots and platform recommendation systems", "unknown", "mixed_unknown", "generation_and_amplification", "confirmed", "observer_and_technical_investigation", "established_evidence", "clm-md-012|clm-md-015|clm-md-017", "Exact models and automation stack are unknown." ], [ "tech-nh-voice", "us-nh-2024-biden-robocall", "ElevenLabs", "AI voice-cloning service", "unknown", "hosted_proprietary", "voice_generation", "confirmed", "creator_reporting", "established_evidence", "clm-nh-001|clm-nh-003", "Provider identification is supported by creator reporting; the FCC orders establish artificial voice but do not name every tool in the production chain." ], [ "tech-brprep-aggregate", "br-2026-election-preparedness", "unknown_or_multiple", "generative AI, recommender and disclosure-compliance systems", "unknown", "mixed_unknown", "prospective_generation_ranking_and_compliance", "regulated_or_monitored_not_case_deployment", "regulatory_scope_and_preparedness_sources", "established_evidence", "clm-br-002|clm-br-003|clm-br-004|clm-br-005|clm-br-006|clm-br-007", "Aggregate preparedness row. The rules establish covered capabilities, providers and response duties, not an exact model, provider stack or specific manipulative deployment." ], [ "tech-br-avatar", "br-2026-bolsonaro-ai-avatar", "unknown", "synthetic image-and-voice avatar system", "unknown", "unknown", "avatar_generation", "confirmed", "content_and_reporting", "established_evidence", "clm-br-010|clm-br-011", "The public record does not identify the model or provider." ], [ "tech-usprep-aggregate", "us-2026-midterms-preparedness", "unknown_or_multiple", "generative AI, synthetic-media, platform and telecom systems", "unknown", "mixed_unknown", "prospective_generation_distribution_and_telecom", "prior_cycle_observed_current_cycle_monitored", "official_risk_assessment_legislative_tracker_and_prior_cycle_records", "established_evidence", "clm-us26-002|clm-us26-003|clm-us26-004|clm-us26-005|clm-us26-006", "Aggregate preparedness row. The evidence establishes regulation, prior-cycle use and monitored capability categories; it does not identify an exact model, provider stack or new 2026 operation." ], [ "tech-us26-ads", "us-2026-campaign-deepfake-ads", "unknown", "campaign video-generation and face/voice synthesis tools", "unknown", "unknown", "synthetic_video_generation", "confirmed", "campaign_artifact_and_reporting", "established_evidence", "clm-us26-008|clm-us26-011|clm-us26-014", "Do not infer a frontier-model provider from output appearance." ], [ "tech-usca11-claude", "us-ca11-2026-wiener-conniechan-chatbot", "Anthropic", "Claude", "unknown", "closed_hosted_product", "interactive_text_generation", "confirmed", "campaign_statement_and_direct_observation", "established_evidence", "clm-usca11-003|clm-usca11-004", "Exact model, API or consumer route, system prompt, retrieval, context and safety settings were not disclosed." ], [ "tech-usca11-stack", "us-ca11-2026-wiener-conniechan-chatbot", "Amazon Web Services|Google", "hosting and analytics stack", "unknown", "commercial_infrastructure", "hosting_analytics_and_data_capture", "reported", "privacy_policy", "established_evidence", "clm-usca11-007", "Infrastructure listing does not establish which service saw which fields." ] ], "content_items": [ [ "item-nh-biden-call", "us-nh-2024-biden-robocall", "AI-cloned Biden robocall", "audio", "Joe Biden", "voter_participation_instruction", "false_authentic_voice", "no", "unauthorised_or_not_affiliated", "not_disclosed_in_call", "deceptive_impersonation", "clm-nh-001|clm-nh-002|clm-nh-005", "The call combined voice clone, targeted list, spoofed caller ID and telecom delivery." ], [ "item-br-bolsonaro-avatar", "br-2026-bolsonaro-ai-avatar", "Jair Bolsonaro campaign-launch avatar", "audiovisual_avatar", "Jair Bolsonaro", "political_endorsement", "synthetic_surrogate", "yes", "unknown", "explicit", "transparent_synthetic_endorsement", "clm-br-010|clm-br-011|clm-br-019", "Disclosure is established; authorisation and legal status remained unresolved at the cutoff." ], [ "item-us26-talarico", "us-2026-campaign-deepfake-ads", "NRSC James Talarico synthetic performance", "video", "James Talarico", "candidate_attack", "fabricated_first_person_performance_using_real_posts", "yes_small_text", "unauthorised_or_not_affiliated", "low_salience", "mixed_quotation_and_fabrication", "clm-us26-008|clm-us26-009|clm-us26-014|clm-us26-015", "Quoted posts were genuine; reaction lines, affect and first-person performance were synthetic." ], [ "item-usca11-chatbot", "us-ca11-2026-wiener-conniechan-chatbot", "ConnieChan.ai interactive parody", "interactive_text_chatbot", "Connie Chan", "candidate_attack_and_issue_framing", "first_person_dynamic_opponent_simulation", "yes", "not_affiliated_or_endorsed", "explicit_site_and_footer", "transparent_satirical_simulation", "clm-usca11-001|clm-usca11-004|clm-usca11-005|clm-usca11-007", "User prompts dynamically affected outputs; private campaign configuration and logs were not available for audit." ] ], "quantitative_observations": [ [ "q-ro24-accounts", "ro-2024-presidential-annulment", "clm-ro-011", "accounts", 25000, "about", "TikTok", "final_two_weeks_before_2024_first_round", "network_size", "unknown", "unknown", "officially_reported", "src-ro-sri-csat-20241204", "Account count is not exposure, unique people, persuasion, or votes." ], [ "q-ro25-accounts", "ro-2025-presidential-repeat", "clm-ro-014", "accounts", 215, "at_least", "Facebook|Instagram|TikTok", "2024-12_to_2025-06-06", "analysed_network_sample", "not_applicable", "no", "researcher_measured", "src-ro-dfrlab-repeat-20250826", "Coordinated or apparently hijacked accounts in the studied set." ], [ "q-ro25-posts", "ro-2025-presidential-repeat", "clm-ro-015", "posts", 8514, "exact_in_sample", "Facebook|Instagram|TikTok", "2024-12_to_2025-06-06", "content_volume", "not_applicable", "no", "researcher_measured", "src-ro-dfrlab-repeat-20250826", "Posts are not users or exposures." ], [ "q-ro25-views", "ro-2025-presidential-repeat", "clm-ro-015", "views", 16000000, "at_least", "Facebook|Instagram|TikTok", "2024-12_to_2025-06-06", "platform_view_events", "unknown", "no", "researcher_aggregated_platform_metrics", "src-ro-dfrlab-repeat-20250826", "Cross-platform views are not unique people and should not be summed with likes." ], [ "q-ro25-likes", "ro-2025-presidential-repeat", "clm-ro-015", "likes", 681000, "at_least", "Facebook|Instagram|TikTok", "2024-12_to_2025-06-06", "engagement_events", "unknown", "no", "researcher_aggregated_platform_metrics", "src-ro-dfrlab-repeat-20250826", "Likes are not exposure or persuasion." ], [ "q-md24-fb-followers", "md-2024-presidential-referendum", "clm-md-014", "followers", 4200, "about", "Facebook", "at_takedown_2024-10-11", "page_followers", "unknown", "no", "platform_reported", "src-md-meta-threat-reporting", "Meta said most followers were outside Moldova or indicated fake engagement." ], [ "q-md24-ig-followers", "md-2024-presidential-referendum", "clm-md-014", "followers", 335000, "about", "Instagram", "at_takedown_2024-10-11", "account_followers", "unknown", "no", "platform_reported", "src-md-meta-threat-reporting", "Meta said most followers were outside Moldova or indicated fake engagement." ], [ "q-md25-accounts", "md-2025-parliamentary-linked-network", "clm-md-015", "accounts", 253, "exact_in_sample", "Facebook|Instagram|TikTok", "study_window", "analysed_network_sample", "not_applicable", "no", "researcher_measured", "src-md-dfrlab-infoleaders-20250924", "Sample, not the full operation." ], [ "q-md25-posts", "md-2025-parliamentary-linked-network", "clm-md-016", "posts", 28708, "exact_in_sample", "Facebook|Instagram|TikTok", "study_window", "content_volume", "not_applicable", "no", "researcher_measured", "src-md-dfrlab-infoleaders-20250924", "Content count, not audience." ], [ "q-md25-views", "md-2025-parliamentary-linked-network", "clm-md-016", "views", 55367950, "exact_in_sample", "Facebook|Instagram|TikTok", "study_window", "platform_view_events", "unknown", "no", "researcher_aggregated_platform_metrics", "src-md-dfrlab-infoleaders-20250924", "Cross-platform view events are non-unique and non-causal." ], [ "q-md25-likes", "md-2025-parliamentary-linked-network", "clm-md-016", "likes", 2207894, "exact_in_sample", "Facebook|Instagram|TikTok", "study_window", "engagement_events", "unknown", "no", "researcher_aggregated_platform_metrics", "src-md-dfrlab-infoleaders-20250924", "Engagement is not belief or vote change." ], [ "q-md25-shares", "md-2025-parliamentary-linked-network", "clm-md-016", "shares", 295608, "exact_in_sample", "Facebook|Instagram|TikTok", "study_window", "engagement_events", "unknown", "no", "researcher_aggregated_platform_metrics", "src-md-dfrlab-infoleaders-20250924", "Engagement is not belief or vote change." ], [ "q-md25-comments", "md-2025-parliamentary-linked-network", "clm-md-016", "comments", 239016, "exact_in_sample", "Facebook|Instagram|TikTok", "study_window", "engagement_events", "unknown", "no", "researcher_aggregated_platform_metrics", "src-md-dfrlab-infoleaders-20250924", "Engagement is not belief or vote change." ], [ "q-nh-attempts", "us-nh-2024-biden-robocall", "clm-nh-004", "call_attempts", 9581, "exact", "telecom", "campaign_window", "delivery_attempts", "unknown", "no", "provider_records_in_enforcement", "src-us-fcc-kramer-final-pdf-20240926", "Attempts are not completed calls, listeners, unique recipients or voters." ], [ "q-nh-completed", "us-nh-2024-biden-robocall", "clm-nh-004", "completed_calls", 3978, "exact", "telecom", "campaign_window", "completed_call_records", "unknown", "no", "provider_records_in_enforcement", "src-us-fcc-kramer-final-pdf-20240926", "Completed calls are not necessarily unique listeners or persuaded voters." ], [ "q-nh-kramer-forfeiture-usd", "us-nh-2024-biden-robocall", "clm-nh-007", "fine_forfeiture_usd", 6000000, "exact", "Federal Communications Commission", "forfeiture_order_2024-09-26", "separate_FCC_forfeiture_against_Steve_Kramer", "not_applicable", "no", "official_enforcement_order", "src-us-fcc-kramer-final-pdf-20240926", "This is the FCC forfeiture against Kramer, not the separate Lingo penalty and not a measure of electoral harm." ], [ "q-nh-lingo-civil-penalty-usd", "us-nh-2024-biden-robocall", "clm-nh-007", "fine_civil_penalty_usd", 1000000, "exact", "Federal Communications Commission", "consent_decree_2024-08-21", "separate_Lingo_Telecom_civil_penalty", "not_applicable", "no", "official_consent_decree", "src-us-fcc-lingo-consent-202408", "This is the separate Lingo Telecom civil penalty and compliance plan, not the Kramer forfeiture and not a measure of electoral harm." ], [ "q-us26-states-deepfake-laws", "us-2026-midterms-preparedness", "clm-us26-002", "states_with_enacted_political_deepfake_laws", 31, "exact", "state_legislation", "snapshot_2026-06-23", "United_States_states", "not_applicable", "no", "legislative_tracker_count", "src-us-ncsl-20260623", "Snapshot count only. Enactment, effective date, scope, prohibitions, disclosure requirements and litigation status differ by state." ], [ "q-us26-reported-synthetic-ads", "us-2026-campaign-deepfake-ads", "clm-us26-010", "synthetic_ads_identified_in_reporting", 3, "at_least", "multiple_campaign_channels", "reporting_through_2026-03-28", "named_ads_in_reviewed_reporting", "not_applicable", "no", "independent_reporting_content_count", "src-us-reuters-midterms-20260328", "A lower-bound content count from reviewed reporting, not a census of campaign ads, impressions, targeting, unique exposure or electoral effects." ], [ "q-usca11-questions", "us-ca11-2026-wiener-conniechan-chatbot", "clm-usca11-006", "submitted_questions", 5058, "exact_campaign_report", "ConnieChan.ai", "first_27_hours", "interaction_events", "unknown", "no", "campaign_reported", "src-us-ca11-mission-20260807", "Not users, conversations, district voters, impressions, accepted answers, or outcomes; repeated and adversarial prompts possible." ], [ "q-usca11-billboards", "us-ca11-2026-wiener-conniechan-chatbot", "clm-usca11-001", "billboards", 2, "exact_campaign_report", "physical", "August_2026", "promotional_placements", "not_applicable", "no", "campaign_reported", "src-us-ca11-sfgate-20260804", "No impression estimate is available." ], [ "q-usca11-retention-policy-days", "us-ca11-2026-wiener-conniechan-chatbot", "clm-usca11-007", "maximum_conversation_retention_days", 90, "policy_maximum", "ConnieChan.ai", "privacy_policy_observed_2026-08-04", "conversation_log_retention_policy", "not_applicable", "no", "reporter_direct_observation_of_privacy_policy", "src-us-ca11-gazetteer-20260804", "The policy allowed retention for up to 90 days; this does not establish that every conversation was retained for 90 days or that staff reviewed any particular log." ] ], "model_evaluations": [ [ "eval-us-aidp-english-2024", "AI Democracy Projects English election-query pilot", "United States", "January 2024", "2024 U.S. election information", "GPT-4-0613|Claude-2 (2023-06-01)|Llama-2-70b-chat-hf|Mixtral-8x7B-Instruct-v0.1|Gemini Pro Preview (2023-12-13)", "versions_reported", "provider APIs and hosted APIs", "en", "model_response", 130, "expert-coded voter-like queries", "accuracy|completeness|bias|harm", "Half of responses were rated inaccurate; GPT-4 performed better on accuracy but had comparable incompleteness.", "Point-in-time domain safety test; not representative of users, all questions, consumer interfaces, future performance, or election effects.", "src-us-proof-aidp-20240227", "public_query_response_ratings", "More than 40 election officials and experts participated; exact row count follows the released study data, while the article emphasises rates." ], [ "eval-us-aidp-spanish-2024", "AI Democracy Projects bilingual Arizona election-query evaluation", "United States", "2024", "Arizona voter information", "Claude 3 Opus|Gemini 1.5 Pro|GPT-4|Llama 3|Mixtral 8x7B v0.1", "versions_reported_to_family_or_release", "APIs", "en|es", "model_response", 250, "25 questions × 2 languages × 5 models", "accuracy|completeness|language_disparity", "52% of Spanish responses and 43% of English responses contained inaccurate information in the reported comparison; 250 responses were dual fact-checked.", "API responses may differ from consumer chatbots; one state and selected questions; no user or electoral-effect measurement.", "src-us-proof-spanish-20241030", "public_prompts_and_ratings", "The article also reports a 48% combined response inaccuracy in prose and a rounded 45% in its summary box; retain source wording and inspect released data for analysis." ], [ "eval-hu-voting-advice-2026", "Hungary 2026 general-purpose AI voting-advice evaluation", "Hungary", "23-24 March 2026", "2026 Hungarian parliamentary election", "ChatGPT Free|Gemini Free", "exact_underlying_versions_unknown", "consumer web interfaces", "hu", "model_output", 200, "5 party profiles × 10 repetitions × 2 systems × 2 prompt conditions", "correct_party_match|consistency|non_running_party_mentions|guidance_despite_disclaimer", "Both systems produced unstable and inaccurate political matching; ChatGPT mentioned Tisza in only 8% of voting-advice responses and both systems frequently mentioned parties not running.", "Structured audit during a live campaign; not a sample of real users and not evidence that outputs affected the decisive result.", "src-hu-liberties-voting-advice-20260721", "archived_outputs_described", "Important closed-model intermediary case with repeated identical prompts and explicit negative electoral-effect finding." ], [ "eval-us-brennan-2026", "Brennan Center election-disinformation capability and safeguard evaluation", "United States", "February-August 2026", "2026 U.S. midterm election-disinformation scenarios", "ChatGPT|Gemini|Grok|Claude|Perplexity|DeepSeek|Meta AI|Runway Gen-4|Flux.2", "multiple_versions_some_unknown", "consumer products", "en|es", "prompt_or_media_test", 14, "multistage chatbot, generation, safeguard-consistency and 14-image detection tests", "conspiracy_rebuttal|generation_acceptance|version_consistency|synthetic_media_detection", "Chatbots generally rebutted established conspiracies, yet all tested generation tools produced deceptive election scenes; ChatGPT, Grok, Claude and Perplexity correctly identified four or fewer of 14 synthetic images, and version-level accept/refuse behaviour was inconsistent.", "Capability and safeguard evidence under researcher tests; not deployment prevalence, audience exposure, or electoral effect. Harmful prompts are intentionally withheld.", "src-us-brennan-eval-20260811", "methodology_available_prompts_restricted_for_safety", "The sample_size field records the explicit image-detection set, not every prompt across the multistage study; consult the methodology for each denominator." ], [ "eval-global-voter-dialogue-rct-2025", "Persuading voters using human-artificial intelligence dialogues", "United States|Canada|Poland", "2024-2025 election periods", "2024 U.S., 2025 Canadian and 2025 Polish national elections", "purpose-built partisan conversational agents", "study_configuration_reported", "experimental chat interface", "en|fr|pl", "participant", 6105, "preregistered randomized experiments during real elections; 2,457 U.S., 1,530 Canadian and 2,118 Polish participants", "attitude_and_vote_intention_change|categorical_vote_choice|factuality", "Disclosed AI dialogues changed reported preferences and intentions in assigned directions, with larger shifts in the Canadian and Polish studies; some factual errors were observed.", "Causal for the experimental conversations and measured self-reports; not observed campaign deployment, actual ballots, aggregate election outcomes, durable effects, or covert manipulation.", "src-global-nature-voter-dialogue-20251204", "paper_and_osf_materials_available", "This is the strongest causal capability bridge in the package and must remain separate from incident prevalence and real-world electoral-effect claims." ], [ "eval-uk-chatbot-voting-advice-rct-2025", "Chatbot Voting Advice Applications inform but seldom sway young unaligned voters", "United Kingdom", "2025", "experimental voting-advice context", "retrieval-augmented voting-advice chatbots", "study_configuration_reported", "experimental chat interface", "en", "experiment", 3, "three randomized experiments grounded in official political documents; participant N must be taken from the paper's study-specific flow", "political_knowledge|candidate_evaluation|vote_preference", "The chatbots improved political knowledge more consistently than downstream candidate evaluations or vote preferences.", "Causal within the experimental samples and interfaces; not a population prevalence estimate or proof that conversational political AI never persuades.", "src-global-pnas-chatbot-voting-advice-20251208", "paper_materials_reported", "sample_size=3 denotes experiments, not participants. Paired with the Nature study to prevent overgeneralizing a single persuasive-chatbot result." ], [ "eval-eu-mistral-voting-assistant-2024", "Mistral and Mixtral as European Parliament voting assistants", "European Union", "2024", "2024 European Parliament election", "Mistral Instruct-v0.2|Mixtral Instruct-v0.1", "open_weight_exact_chat_versions_reported", "research evaluation with and without curated retrieval", "en", "party_question_pair", 810, "30 questions × 27 parties, evaluated in four contextual settings", "answer_accuracy|group_disparity|retrieval_gain", "Mixtral averaged about 82% accuracy, with substantial group-level variation (roughly 50-95%); expert-curated context improved performance by about nine percentage points.", "Point-in-time model audit; not public deployment, user behavior or election effect.", "src-eu-emnlp-voting-assistant-202411", "paper_and_code_links_available", "sample_size is the 810 party-question pairs before repeated contextual settings. Provides an open-weight comparator; no manipulatory election deployment is inferred from benchmark performance." ] ], "control_shift_assessments": [ [ "ata-ro24", "ro-2024-presidential-annulment", "voter_attention_and_electoral_choice", "Romanian voters' attention and electoral choice", "campaign-network operators and platform ranking systems", "opaque promotion, coordinated accounts, and recommender amplification", "network sponsorship and ranking logic partly opaque", "low_or_unknown", "low_during_campaign_high_after_institutional_response", "campaign_period_with_institutional_aftereffects", "campaign-network operators and platform ranking systems", "unequal agenda-setting, impaired informed choice, and institutional trust shock", "Opaque coordinated promotion and an institutional annulment are observed as separate facts; individual behaviour and a causal electoral effect are not.", "observed", "observed", "insufficient_evidence", "not_observed", "not_observed", "control_mechanism_observed", "If the coordination, sponsorship, or distribution-control evidence fails independent verification, the control-mechanism hypothesis is not supported; an agency-change conclusion also requires measured individual outcomes.", "Comparable exposed and unexposed voters, or opaque and transparently attributed campaign content, with pre-specified attention, belief, and choice-autonomy outcomes.", "Official and investigative records describe opaque promotion, coordinated accounts, platform distribution, and the court's annulment decision.", "The coded mechanism could shift agenda-setting control toward operators and rankers; the dataset does not infer that this changed a person's agency or caused annulment.", "observed_mechanism", "clm-ro-002|clm-ro-003|clm-ro-006|clm-ro-007|clm-ro-013|clm-ro-020", "Generative AI was not established as material. The court decision is an institutional outcome, not evidence of a voter-level agency change or causal vote effect." ], [ "ata-ro25", "ro-2025-presidential-repeat", "authorship_attention_and_local_identity", "Romanian users and authentic local speakers", "unidentified network operator", "synthetic personas, apparently hijacked accounts, and coordinated cross-platform amplification", "operator and provenance opaque", "low_or_unknown", "medium_after_takedown", "months", "unidentified network operator and platform distribution systems", "manufactured local consensus and controlled agenda cues", "A coordinated network, content, and non-unique platform activity are observed; belief, choice, turnout, and election effects are not.", "observed", "observed", "insufficient_evidence", "not_observed", "not_observed", "control_mechanism_observed", "If the accounts are shown to be independent, authentically local, or not centrally coordinated, the control-mechanism hypothesis is not supported; reach metrics alone cannot establish agency change.", "Users exposed to coordinated synthetic-persona content versus matched users exposed to authentic, independently produced local content, with provenance comprehension and decision outcomes measured.", "Researchers documented at least 215 coordinated or apparently hijacked accounts, synthetic personas, content, and aggregated platform activity.", "The network could control apparent local identity and salience; no operator-level intent to displace agency and no user-level agency change are directly observed.", "observed_mechanism", "clm-ro-014|clm-ro-015|clm-ro-016|clm-ro-017|clm-ro-018|clm-ro-021", "Pro-Russia narratives do not establish Russian-state command. The repeat election and the 2024 annulment are analytically separate from this network record." ], [ "ata-md24", "md-2024-presidential-referendum", "preference_participation_and_trust", "Moldovan voters", "Shor-linked and Russia-linked influence and finance networks", "payments, coordinated tasking, bots, spoofed media, and synthetic content", "sponsor, finance, and message provenance partly opaque", "low_or_unknown", "medium_after_enforcement", "multi_election", "foreign-linked political-finance and influence networks", "covert inducement and manufactured social proof could impair independent deliberation", "Payment and influence infrastructure, an inducement attempt, and some payment acceptance are documented; how recipients voted and the electoral effect are unknown.", "observed", "observed", "observed", "not_observed", "not_observed", "attempted_transfer_observed", "If payments were not conditioned, tasked, or linked to electoral conduct, the attempted-transfer coding fails; accepted payment alone cannot establish changed preference or vote.", "Credibly identified recipients and comparable non-recipients under a design addressing selection, with task compliance, belief, turnout, and vote outcomes measured separately.", "Official, observer, platform, and reporting records document coordinated influence infrastructure and allegations/evidence of electoral inducement; AI is one component.", "The covert inducement and impersonation mechanisms constitute an observed attempt to influence decision control, but recipient agency change and votes changed are not observed.", "observed_attempt_with_unmeasured_agency", "clm-md-003|clm-md-004|clm-md-005|clm-md-009|clm-md-010|clm-md-011|clm-md-020", "The removed 138,000 mixed accounts-and-transactions aggregate is not a valid reach observation. Administrative advantage and media imbalance remain competing explanations and separate harms." ], [ "ata-md25", "md-2025-parliamentary-linked-network", "authorship_agenda_and_peer_signal", "Moldovan social-media users and authentic local speakers", "InfoLeaders curators and linked political networks", "central scripts locally adapted by paid participants and reciprocal amplification", "coordination and payment opaque to audiences", "low_or_unknown", "medium_after_disruption", "months", "network curators and financiers", "paid coordination can masquerade as peer speech and control apparent local salience", "Paid tasking, coordinated accounts, content, and engagement are observed; persuasion, turnout, votes, and seats are not.", "observed", "observed", "insufficient_evidence", "not_observed", "not_observed", "control_mechanism_observed", "If tasking, payment, or reciprocal amplification is not reproducible at account level, the control-mechanism coding fails; engagement totals cannot establish an agency outcome.", "Matched users or content streams exposed to centrally tasked paid personas versus authentic independent speakers, with provenance recognition and decision outcomes measured.", "A documented paid InfoLeaders sample used central tasking, local personas, and reciprocal engagement; platform totals are non-unique.", "The mechanism could shift control over apparent peer signals toward curators; no individual agency change or electoral effect is observed.", "observed_mechanism", "clm-md-012|clm-md-015|clm-md-016|clm-md-017|clm-md-018|clm-md-019|clm-md-021", "Components with different attribution and measurement scopes must not be collapsed into one operator or reach total." ], [ "ata-nh", "us-nh-2024-biden-robocall", "authorship_relevance_and_participation_instruction", "New Hampshire Democratic primary recipients", "Steve Kramer and the telecom delivery chain", "cloned presidential voice, politically selected call list, and spoofed caller ID", "high apparent-authorship deception", "medium", "high_after_enforcement", "short_campaign_burst", "political operator and telecom distribution chain", "fabricated authoritative instruction could displace an informed participation decision", "The impersonated demobilising instruction, operator, call attempts, and completed calls are observed; listening, belief, action, and turnout effect are not.", "observed", "observed", "observed", "not_observed", "not_observed", "attempted_transfer_observed", "If the audio was not synthetic, the instruction was not demobilising, or the attributed operator did not select and distribute it, attempted-transfer coding fails; delivery counts alone do not establish agency change.", "Recipients versus matched non-recipients, or completed versus failed delivery under a credible design, measuring listening, belief, action, and turnout without treating calls as unique voters.", "An admitted operator created an AI-cloned Biden voice telling recipients not to vote in the primary; FCC records document the delivery chain and call counts.", "The impersonated participation instruction is an observed attempt to displace decision control; no recipient-level agency change is observed.", "observed_attempt_with_unmeasured_agency", "clm-nh-001|clm-nh-002|clm-nh-004|clm-nh-005|clm-nh-009|clm-nh-010|clm-nh-012", "Criminal acquittal, FCC findings, and civil default are distinct. None supplies evidence of belief, action, turnout, or votes changed." ], [ "ata-brprep", "br-2026-election-preparedness", "future_information_and_platform_governance", "not_applicable", "not_applicable", "prospective monitoring of synthetic media and provider outputs", "prospective", "not_assessed", "not_assessed", "future_election", "not_applicable", "prospective risk only", "Election rules, provider obligations, and institutional preparations are observed; no incident or agency outcome is coded.", "not_assessed", "not_observed", "not_applicable", "not_applicable", "not_applicable", "not_an_incident", "Not applicable unless a discrete incident is promoted with claim-level evidence and an explicit unit of analysis.", "No incident comparator; future promoted records require an incident-specific comparator.", "TSE rules, compliance obligations, partnerships, and monitoring arrangements.", "", "not_applicable", "clm-br-002|clm-br-003|clm-br-004|clm-br-006|clm-br-007|clm-br-009", "Preparedness file; excluded from incident counts and from agency conclusions." ], [ "ata-bravatar", "br-2026-bolsonaro-ai-avatar", "political_representation_and_audience_interpretation", "Jair Bolsonaro as represented person and voters evaluating the endorsement", "campaign producer controlling the synthetic representation", "disclosed synthetic political surrogate", "AI generation disclosed; authorisation unresolved", "low", "medium_after_disclosure_and_legal_challenge", "single_event_with_reuse_risk", "campaign producer controlling the synthetic representation", "unauthorised synthetic representation could appropriate political voice; disclosure may preserve audience verification", "The avatar, message, and AI disclosure are observed; authorisation, audience interpretation, belief, and electoral effects are unresolved.", "observed", "observed", "insufficient_evidence", "not_observed", "not_observed", "insufficient_evidence", "If authorisation is established and audiences reliably understand the disclosed surrogate as campaign-authored representation, the appropriation hypothesis weakens; agency change still requires measured audience outcomes.", "Authorised versus unauthorised and prominently disclosed versus undisclosed synthetic surrogates, measuring attribution accuracy, perceived endorsement, and decision autonomy.", "A disclosed synthetic Bolsonaro avatar delivered an endorsement at a campaign launch; legal claims and a defence were filed, but final merits resolution and authorisation were not identified.", "The producer controlled the synthetic representation, but whether that displaced or preserved anyone's agency is an untested boundary hypothesis.", "boundary_case", "clm-br-010|clm-br-011|clm-br-014|clm-br-017|clm-br-019|clm-br-020", "Transparent disclosure is material counterevidence. Disclosure alone does not prove agency preservation, and production control alone does not prove agency transfer." ], [ "ata-usprep", "us-2026-midterms-preparedness", "future_information_and_election_trust", "not_applicable", "not_applicable", "prospective monitoring of campaign, foreign, platform, telecom, and AI-provider mechanisms", "prospective and fragmented governance", "not_assessed", "not_assessed", "future_election", "not_applicable", "prospective risk only", "Election timing, fragmented law, and prior-cycle risk assessments are observed; no new 2026 operation or agency outcome is coded.", "not_assessed", "not_observed", "not_applicable", "not_applicable", "not_applicable", "not_an_incident", "Not applicable unless a discrete 2026 incident is promoted with claim-level evidence and an explicit unit of analysis.", "No incident comparator; future promoted records require an incident-specific comparator.", "Scheduled election context, enacted state laws, federal governance gaps, and prior-cycle assessments.", "", "not_applicable", "clm-us26-001|clm-us26-002|clm-us26-003|clm-us26-004|clm-us26-005|clm-us26-007", "Preparedness file; excluded from incident counts and from agency conclusions." ], [ "ata-usads", "us-2026-campaign-deepfake-ads", "candidate_voice_face_and_affect", "represented candidates and voters interpreting campaign evidence", "campaign sponsors producing synthetic first-person performances", "synthetic candidate simulations with variable disclosure and quotation fidelity", "disclosure varies and can be low-salience", "low", "medium", "campaign_cycle", "campaign sponsors producing and distributing synthetic representations", "synthetic performance can make the sponsor control apparent speech, timing, image, and affect", "Multiple attributed synthetic assets are observed; comparable exposure, attribution error, belief, and election effects are not.", "observed", "observed", "insufficient_evidence", "not_observed", "not_observed", "control_mechanism_observed", "If the synthetic elements are obvious, accurately attributed, faithful to genuine quotations, and do not alter audience attribution or interpretation, the stronger control-shift hypothesis weakens; content production alone cannot establish agency change.", "Authentic footage versus labelled and unlabelled synthetic performances, stratified by quotation fidelity, with attribution accuracy and decision outcomes measured.", "Named campaign sponsors released synthetic candidate performances; disclosure and fidelity varied across assets.", "Sponsors controlled simulated voice, face, timing, and affect, but whether this displaced voter or candidate agency is untested.", "observed_mechanism", "clm-us26-008|clm-us26-009|clm-us26-010|clm-us26-011|clm-us26-012|clm-us26-013|clm-us26-014|clm-us26-016", "Some assets use genuine quotations, labels, or obvious satire. Content-level distinctions are required before any stronger inference." ], [ "ata-usca11", "us-ca11-2026-wiener-conniechan-chatbot", "representation_agenda_query_privacy_and_candidate_evaluation", "Connie Chan as represented person and users asking questions", "Scott Wiener for Congress and its provider stack", "disclosed campaign-controlled first-person opponent simulation with policy-permitted query-log review", "AI and parody disclosed; system prompt, retrieval, actual log use, and guardrails opaque", "high", "high_after_removal", "four_day_deployment_plus_90_day_retention_policy", "campaign deployer; secondarily model and infrastructure providers", "sponsor control over opponent simulation and query data could shape representation and agenda", "Deployment, model family, question count, privacy-policy terms, disclosure, and removal are observed; actual retention, unique users, interpretation, and electoral effects are not.", "observed", "observed", "insufficient_evidence", "not_observed", "not_observed", "insufficient_evidence", "If users reliably understand the tool as sponsor-authored parody, outputs are traceably grounded, logs are not used beyond disclosed purposes, and no decision-relevant difference appears against an ordinary attack ad, the stronger control-shift hypothesis weakens.", "Disclosed parody chatbot versus ordinary campaign attack material and versus an undisclosed first-person simulation, measuring attribution, source verification, perceived candidate position, and decision autonomy.", "The campaign commissioned and disclosed a Claude-powered parody, controlled its presentation, permitted limited log review in policy, reported question volume, and removed the tool after criticism.", "The system could shift representational and evidentiary control toward the sponsor, but explicit parody and absent user-outcome evidence make this a boundary hypothesis.", "boundary_case", "clm-usca11-002|clm-usca11-003|clm-usca11-004|clm-usca11-005|clm-usca11-006|clm-usca11-007|clm-usca11-009|clm-usca11-010|clm-usca11-011", "Explicit parody and disclosure are essential counterevidence. Neither question count nor campaign control demonstrates a change in user agency." ] ] }