case_id,case_family_id,title,country,iso3,region,election_name,election_type,election_date_start,election_date_end,case_status,as_of_date,anchor_case,record_type,manipulation_assessment,summary,actor_controlling_system,actor_type,actor_attribution_status,target_population,influence_vector,platforms,mechanism_class,ai_role,generative_ai_status,affected_state,control_shift_hypothesis,hypothesized_power_recipient,democratic_harm,occurrence_status,reach_status,reach_estimate,reach_unit,behavioural_effect_status,electoral_effect_status,institutional_outcome,response,inclusion_tier,counterevidence,open_questions,last_verified
ro-2024-presidential-annulment,romania-presidential-2024-2025,Romania 2024 presidential election: coordinated digital campaign and annulment,Romania,ROU,Europe,2024 Romanian presidential election,presidential,2024-11-24,2024-11-24,retrospective,2026-08-11,true,election_wide_case,mixed_documented_manipulation,"Romania's Constitutional Court annulled the presidential election after the first round amid officially documented concerns about campaign irregularities, opaque coordinated promotion, platform recommender risks, cyber activity, and foreign interference. Generative AI is not established as a material driver.",mixed and partly unresolved campaign networks; platform recommender systems; alleged foreign-linked actors,campaign_network|platform|alleged_foreign_actor,mixed,Romanian voters,"short-form video recommendations, coordinated accounts, undeclared promotion, cyber and information operations",TikTok|Telegram|Facebook|web,recommender_amplification|coordinated_inauthentic_behaviour|paid_influence|institutional_response,material_amplification,not_established,attention|belief|trust|electoral choice,"Hypothesis: opaque promotion and recommender amplification could shift control over agenda-setting from voters toward campaign-network operators and ranking systems. The record observes a mechanism and institutional response, not individual agency change or a causal link to annulment.",campaign-network operators and platform ranking systems,"Documented unequal campaign conditions, opacity, and institutional disruption; impaired informed choice and causal effects on votes remain unmeasured.",partly_confirmed,platform_reported,,,unknown,unknown,The Constitutional Court annulled the entire presidential electoral process on 6 December 2024.,Court annulment; declassification; European Commission DSA proceedings; platform removals; repeat election.,A,"The precise sponsor, foreign attribution, and causal contribution of platform dynamics remain contested; some reporting indicates a domestic party-financed campaign may have been repurposed.",Who financed and directed each network component? What organic versus coordinated reach occurred? Did any exposure change voting behaviour? What will the DSA proceeding finally establish?,2026-08-11
ro-2025-presidential-repeat,romania-presidential-2024-2025,Romania 2025 repeat presidential election and cross-platform synthetic-persona network,Romania,ROU,Europe,2025 Romanian presidential election repeat,presidential,2025-05-04,2025-05-18,retrospective,2026-08-11,true,observed_network,manipulation_confirmed,"The repeat election is coded as the institutional aftermath of the annulment and as a distinct cross-platform influence file. DFRLab documented at least 215 coordinated or apparently hijacked accounts, synthetic Romanian personas, generative-AI assets, and a pivot from support for Călin Georgescu to George Simion. The operation's sponsor and any voter effect remain unknown.",unknown cross-platform network operator; platform distribution systems; responding election authorities,unknown_network|platform|institution,unknown,Romanian voters,"coordinated and hijacked accounts, synthetic personas, AI avatars, false local-news identity, cross-platform hashtags, and institutional decisions",TikTok|Facebook|Telegram|web,institutional_response|recommender_amplification|coordinated_inauthentic_behaviour,mixed,confirmed,trust|participation|belief,Hypothesis: synthetic personas and coordinated amplification could let an unidentified operator shape apparent local consensus. Network activity is observed; no individual agency change or electoral effect is measured.,unidentified network operator and platform distribution systems,"Potential distrust, polarisation, and uncertainty around electoral legitimacy; no individual trust or agency outcome is measured.",confirmed,measured,16000000,non-unique cross-platform views reported for the studied network,unknown,unknown,A repeat presidential election was held in May 2025.,"Expanded observation, platform engagement, partial deplatforming, and election-integrity safeguards.",A,"The 16 million figure is a sum of platform views, not unique Romanian voters or authenticated human exposure. DFRLab described the network as pro-Russia from its narratives, but could not identify the operator or prove Russian-state command. A completed repeat election does not resolve the disputed causal basis of the 2024 annulment.",Who operated and financed the network? How much reach was authentic and in-country? Which safeguards measurably reduced manipulation? How did the annulment itself affect trust and participation?,2026-08-11
md-2024-presidential-referendum,moldova-elections-2024-2025,Moldova 2024 presidential election and EU referendum: hybrid interference network,Moldova,MDA,Europe,2024 presidential election and EU constitutional referendum,presidential|referendum,2024-10-20,2024-11-03,retrospective,2026-08-11,true,election_wide_case,mixed_documented_manipulation,"A high-evidence hybrid-interference file combining alleged illicit foreign finance and vote buying with disinformation, spoofed media, coordinated accounts, bots, and some AI-generated content. AI is one layer of the operation, not a demonstrated causal explanation for the narrow referendum result.",Ilan Shor-linked network and alleged Russian sponsors; coordinated account operators; platforms,political_network|alleged_foreign_actor|platform,officially_attributed,"Moldovan voters, including linguistic minorities and diaspora","payments, messaging apps, spoofed media, bots, coordinated accounts, synthetic video",Telegram|Facebook|Instagram|TikTok|YouTube|web,illicit_finance|coordinated_inauthentic_behaviour|bot_amplification|spoofed_media|synthetic_video,mixed,confirmed,belief|trust|preference|participation,"Hypothesis: covert finance, inducement, coordinated amplification, and impersonation could shift control over electoral choice toward sponsors and operators. An inducement attempt and some payment acceptance are documented; how recipients voted is unknown.",foreign-linked political-finance and influence networks,"Documented covert-inducement and unequal-campaign-condition concerns; impaired informed choice and institutional distrust are plausible harms, not measured individual outcomes.",partly_confirmed,unknown,,,unknown,unknown,The referendum passed and Maia Sandu won the presidential runoff; investigations and platform takedowns continued.,Police and prosecutorial investigations; account and channel removals; campaign-finance enforcement; international observation and public attribution.,A,Russia and Ilan Shor denied key allegations; official recipient counts do not establish how every payment affected voting; observers also identified incumbent advantage and misuse of public resources.,"How many recipients acted on payments? Which synthetic pieces achieved material exposure? What portion of the network was centrally directed? What vote effect, if any, occurred?",2026-08-11
md-2025-parliamentary-linked-network,moldova-elections-2024-2025,Moldova 2025 parliamentary election: continuation of hybrid influence infrastructure,Moldova,MDA,Europe,2025 Moldovan parliamentary election,parliamentary,2025-09-28,2025-09-28,retrospective,2026-08-11,false,observed_network,manipulation_confirmed,"Linked follow-on evidence shows persistence and adaptation of the 2024 influence infrastructure. A documented paid InfoLeaders network combined inauthentic local personas, Telegram tasking, reciprocal engagement and AI-assisted message adaptation; separate records described synthetic/staged media and context-adaptive bots. Components must not be collapsed into one operator or reach total.",foreign-funded political and influence networks,political_network|alleged_foreign_actor,mixed,Moldovan voters,"social platforms, messaging apps, illicit finance, synthetic and spoofed media",Telegram|Facebook|Instagram|TikTok|YouTube|web,illicit_finance|coordinated_inauthentic_behaviour|bot_amplification|synthetic_video,mixed,confirmed,belief|trust|preference,"Hypothesis: centrally tasked, paid local personas and reciprocal amplification could shift agenda control toward network curators. The mechanism is observed; user belief, choice, and vote effects are not.",foreign-funded influence-network curators and financiers,Potential unequal campaign conditions and impaired informed choice; neither individual agency change nor an electoral effect is measured.,partly_confirmed,measured,55367950,non-unique cross-platform views reported for the InfoLeaders sample,unknown,unknown,Election completed under an intensified enforcement and information-integrity response.,"Network disruption, financial investigations, platform actions, public warnings, and election observation.",A,The platform totals are not unique people or authenticated Moldovan voters. Operational ties to Shor-affiliated or Russia-linked structures vary by component; direct Kremlin command is not proved for every asset. The presence of AI content and automated accounts does not establish decisive reach or electoral effect.,"Which 2024 network components persisted, which were newly created, and which shared command? What was authentic in-country exposure? Which interventions worked?",2026-08-11
us-nh-2024-biden-robocall,new-hampshire-biden-robocall-2024,New Hampshire 2024 AI-cloned Biden robocall,United States,USA,North America,2024 New Hampshire Democratic presidential primary,presidential_primary,2024-01-23,2024-01-23,retrospective,2026-08-11,true,observed_incident,manipulation_confirmed,"An AI-cloned voice impersonating President Joe Biden was distributed by spoofed robocalls shortly before the New Hampshire primary and told recipients not to vote in that contest. The operator admitted creating the call, the FCC imposed civil penalties, and a state jury later acquitted him of all criminal charges.",Steve Kramer and telecom distribution providers,political_consultant|telecom_provider,admitted,New Hampshire Democratic primary voters,"AI voice cloning, caller-ID spoofing, mass robocalls",telephone network,voice_clone|telecom_robocall|voter_suppression,core_generation,confirmed,belief|participation,"Hypothesis: a cloned authoritative voice, spoofed caller ID, and selected call list could displace recipients' control over a participation decision. A demobilising attempt and call delivery are observed; belief or action is not.",political operator and telecom distribution chain,"Observed impersonation of authority and attempted demobilisation; reduced trust is a plausible harm, not a measured outcome.",confirmed,measured,9581,call attempts initiated; 3978 completed calls carried by Lingo according to FCC records,unknown,unknown,FCC $6 million forfeiture; Lingo $1 million consent decree and compliance plan; state criminal acquittal; later federal civil default judgment and injunction; legal clarification that AI-generated voices fall within TCPA artificial-voice restrictions.,"State investigation and charges; FCC declaratory ruling, enforcement, and provider settlement.",A,"The jury acquitted Kramer of all 22 state criminal charges; the FCC finding, civil default, and acquittal are distinct proceedings. No credible evidence quantifies changed turnout or votes.","How many recipients listened, believed, or acted on the call? How should liability be allocated across creator, intermediary, and originating provider?",2026-08-11
br-2026-election-preparedness,brazil-general-2026,Brazil 2026 general election: AI governance and integrity preparedness,Brazil,BRA,South America,2026 Brazilian general election,presidential|legislative|gubernatorial,2026-10-04,2026-10-25,ongoing,2026-08-11,true,preparedness_file,preparedness_not_incident,"A prospective monitoring file grounded in enacted TSE rules, platform-compliance obligations, institutional partnerships, and observed 2026 incidents. It does not predict electoral harm.","campaigns, AI providers, platforms, election authorities, and unidentified influence actors",mixed,unknown,Brazilian voters,"synthetic media, recommender outputs, political advertising, bots, voice cloning, platform distribution",major social platforms|generative AI systems|messaging apps|web,synthetic_video|voice_clone|recommender_amplification|bot_amplification|institutional_response,mixed,confirmed,belief|trust|preference|participation,Monitoring hypothesis only: synthetic generation and ranking systems could alter control over voters' information environments. No incident or agency change is asserted.,"potential campaigns, AI providers, and large platforms, constrained by election authorities","prospective: deception, impersonation, discriminatory political violence, suppression, and loss of confidence",partly_confirmed,unknown,,,unknown,unknown,Pending; first round scheduled for 4 October 2026 and runoff for 25 October 2026.,"Mandatory AI disclosure, time-limited ban on new candidate/public-figure synthetic media, provider duties, complaint channels, compliance plans, expert partnerships, and monitoring council.",A,Brazil's 2024 municipal election research found far less high-impact AI abuse than feared; preparedness and incident counts do not imply electoral effect.,Will providers comply in practice? How quickly will labelled and unlabelled content be detected and adjudicated? Will rules apply consistently across closed messaging and AI assistants?,2026-08-11
br-2026-bolsonaro-ai-avatar,brazil-general-2026,Brazil 2026 Bolsonaro AI avatar at Flávio Bolsonaro campaign launch,Brazil,BRA,South America,2026 Brazilian presidential election,presidential,2026-10-04,2026-10-25,ongoing,2026-08-11,true,observed_incident,transparent_contested_use,A disclosed AI avatar of former president Jair Bolsonaro appeared at the launch of his son Flávio's presidential campaign. Federação Brasil da Esperança challenged it in TSE representation RP 0601315-97.2026.6.00.0000; the defence argued that disclosure and the absence of deceptive editing made it lawful. The TSE had not finally resolved the dispute as of the cutoff.,Flávio Bolsonaro campaign or allied producer; authorisation by Jair Bolsonaro unresolved,campaign,credibly_reported,Brazilian voters and campaign attendees,disclosed synthetic image and voice presented as an endorsement,campaign event|social video|news media,synthetic_video|voice_clone,core_generation,confirmed,preference|perceived endorsement|trust,Boundary hypothesis: a disclosed synthetic surrogate could shift control over a represented person's public voice toward a campaign producer. Disclosure is observed; authorisation and any user agency change are unresolved.,campaign producer controlling the synthetic representation,contested: possible circumvention of legal restrictions and manipulation through synthetic political presence,confirmed,qualitative,,,unknown,unknown,TSE representation RP 0601315-97.2026.6.00.0000 pending on the merits at the dataset cutoff; an internal TSE meeting was scheduled for 13 August 2026.,Court petitions and public legal scrutiny under TSE synthetic-content rules.,A,The avatar disclosed that it was AI-generated; disclosure weakens a simple covert-deception theory and the campaign denied that the item was an unlawful deepfake. Jair Bolsonaro's authorisation was denied by his lawyers and remains unknown.,Who authorised and scripted the avatar? Does a labelled clone circumvent communication restrictions? How will TSE distinguish transparent simulation from manipulative synthetic representation?,2026-08-11
us-2026-midterms-preparedness,us-midterms-2026,United States 2026 midterm election: fragmented AI governance and preparedness,United States,USA,North America,2026 United States midterm election,legislative|state|local,2026-11-03,2026-11-03,ongoing,2026-08-11,true,preparedness_file,preparedness_not_incident,"An ongoing monitoring file for a federal election governed by a state-law patchwork, narrow federal campaign-law tools, existing telecom rules, voluntary platform policies, and documented campaign use of synthetic media.","campaigns, political committees, foreign influence actors, platforms, telecom and AI providers",mixed,mixed,United States voters,"synthetic ads, voice and video impersonation, recommender distribution, robocalls, coordinated influence",television|social platforms|telephone network|generative AI systems|web,synthetic_video|voice_clone|telecom_robocall|recommender_amplification|coordinated_inauthentic_behaviour|institutional_response,mixed,confirmed,belief|trust|preference|participation,Monitoring hypothesis only: synthetic production and fragmented distribution governance could alter control over voters' information environments. No 2026 incident or agency change is asserted in this preparedness record.,"potential campaign committees, platforms, and infrastructure providers","Prospective risk categories include fabricated candidate speech, disclosure ambiguity, suppression attempts, and erosion of epistemic trust; this preparedness record does not establish a 2026 incident or outcome.",partly_confirmed,qualitative,,,unknown,unknown,Pending; general election scheduled for 3 November 2026.,State disclosure and prohibition laws; FCC artificial-voice enforcement; election-security guidance; platform policies; fact-checking and campaign responses.,A,Generative AI amplified existing risks more often than creating wholly new ones in prior cycles; observed synthetic ads do not establish electoral effects.,How will state rules apply across jurisdictions? Will disclosures be salient? Which foreign or domestic networks will achieve measurable reach? Can authentic communications remain verifiable?,2026-08-11
us-2026-campaign-deepfake-ads,us-midterms-2026,United States 2026 campaign-produced synthetic candidate ads,United States,USA,North America,2026 United States midterm campaigns,legislative|state,2026-11-03,2026-11-03,ongoing,2026-08-11,true,observed_campaign_set,mixed_documented_manipulation,"By March 2026, campaigns and national committees had deployed realistic AI-generated ads depicting opponents saying or doing things they had not recorded, sometimes with small or low-salience disclosures. The 11 March NRSC Talarico ad converted genuine old posts into a synthetic performance and added reaction lines not present in those posts. The case is observed and attributed, but its behavioural or electoral effect is unknown.",National Republican Senatorial Committee and individual campaigns; additional actors may be added,campaign|political_committee,admitted,voters in contested federal and state races,AI-generated political video distributed as campaign advertising and social content,social video|digital advertising|web,synthetic_video|paid_influence,core_generation,confirmed,belief|candidate evaluation|trust,Hypothesis: simulated first-person performance can shift control over a candidate's audiovisual representation toward a sponsor. The assets are observed; any change in voter agency or behaviour is not.,campaign sponsors producing and distributing synthetic representations,"possible misattribution of speech, confusion, normalisation of fabricated evidence, and erosion of trust",confirmed,qualitative,,,unknown,unknown,Pending.,"Candidate objections, voluntary or state-required disclosures, media verification, and calls for clearer rules.",A,"Some ads reproduce genuine prior words, contain disclosures, or are framed as satire; these features complicate classification as deception.",Were disclosures noticed? What impressions and targeting occurred? Do viewers distinguish quotation from fabricated audiovisual evidence? What legal rules apply in each state?,2026-08-11
us-ca11-2026-wiener-conniechan-chatbot,us-midterms-2026,California 11th District: campaign-controlled Connie Chan parody chatbot,United States,USA,North America,2026 California 11th congressional district general election,legislative,2026-11-03,2026-11-03,ongoing,2026-08-12,false,observed_incident,transparent_contested_use,"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.",Scott Wiener for Congress campaign; technical constructor unidentified,campaign|unidentified_vendor,admitted,California 11th District voters and politically interested web visitors; Connie Chan as represented person,"interactive first-person opponent simulation, campaign-selected issue framing, web interface, billboards, and earned media",web|physical_billboards|Claude,synthetic_text|interactive_impersonation|campaign_attack|data_capture,core_generation,confirmed,candidate evaluation|agenda-setting|perceived representation|trust|query privacy,"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.",campaign deployer controlling the interface and framing; secondarily model and infrastructure providers,"Observed first-person opponent simulation and public controversy over racialised and gendered outputs; misrepresentation and normalisation harms remain hypotheses, and electoral effect is unknown.",confirmed,campaign_reported_nonunique_questions,5058,submitted_questions_in_first_27_hours,unknown,unknown,No adjudication or regulatory outcome was identified by the cutoff.,"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.",A,"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.","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?",2026-08-12