Metatron Intelligence
Today: the five AI headlines of the day → and the AI Wiki
What a domain is How to read this page

One of twenty-four connected subject areas

Domains are how claims relate to one another inside the Atlas. Desks are how readers find them. A claim can sit in more than one domain without being counted twice.

Coverage here is not importance

A domain with more claims has been worked on more, which is a fact about this publication rather than a fact about the world.

Everything returns to the record

Each entry links to its canonical claim record, with its source, its non-claim and its update condition intact.

New to this publication?

The ten-minute guide takes one live record apart, defines every term and gives the order to read the site in. Start here →

Horizontal · D16

Creative Technology & Synthetic Media

320 connected claims · 230 source-domain links · Priority P1

Search and filter this domain →
  1. MWITA-CR-2026-0012Generative Medien senken Produktions- und Lokalisierungskosten, erhöhen aber eine Herkunfts- und Rechteschuld, weil Persona-, Asset-, Trainings-/Lizenz- und Disclosure-Rechte explizit nachweisbar werden müssen.Evidence B · P1
  2. MWITA-CR-2026-0026Lokale Sprache ist Infrastruktur: SEA-LION, BHASHINI, Masakhane und Latam-GPT adressieren Daten-, Benchmark-, Code-Switching-, Rechte- und Repräsentationslücken, die eine nachträgliche Übersetzung nicht zuverlässig löst.Evidence B · P0
  3. MWITA-CD-2026-024From 2 August 2026, EU AI Act Article 50 generally requires providers of AI systems generating synthetic audio, image, video or text to mark outputs in a machine-readable format and make them detectable as artificially generated or manipulated.Evidence A · P0
  4. MWITA-CD-2026-025From 2 August 2026, EU professional deployers must disclose deepfake content, subject to Article 50’s role and context provisions.Evidence A · P0
  5. MWITA-CD-2026-026EU AI Act Article 50 provides an adapted disclosure approach for evidently artistic, creative, satirical or fictional works so disclosure does not hamper display or enjoyment.Evidence A · P1
  6. MWITA-CD-2026-027China’s AI-generated synthetic-content labeling measures, effective 1 September 2025, distinguish explicit visible or audible labels from implicit labels.Evidence A · P0
  7. MWITA-CD-2026-028China’s labeling measures require implicit metadata to include the content attribute, provider identifier and content identifier.Evidence A · P0
  8. MWITA-CD-2026-029China’s labeling measures require internet information distribution services to inspect synthetic-content metadata and apply the prescribed notices or labels.Evidence A · P1
  9. MWITA-CD-2026-030In November 2025, the CAC reported corrective actions against apps for missing explicit and implicit AI-content labels, including interviews, correction orders, removals and deactivations.Evidence A · P1
  10. MWITA-CD-2026-031India’s IT Rules amended on 10 February 2026 require permitted synthetic information to carry a prominent visible or audible label.Evidence A · P0
  11. MWITA-CD-2026-032India’s 2026 IT Rules require, where technically feasible, persistent metadata or provenance for permitted synthetic information, including a unique resource identifier, and prohibit enabling removal or suppression of the label.Evidence A · P0
  12. MWITA-CD-2026-033C2PA 2.2 specifies signed manifests, assertions or attestations, soft bindings and related security and user-experience guidance.Evidence A · P1
  13. MWITA-CD-2026-034C2PA 2.2 does not rate provenance data as good or bad; a valid provenance chain is not a semantic truth judgment.Evidence A · P0
  14. MWITA-CD-2026-035A 2026 independent formal-analysis preprint reports that the C2PA specifications it examined did not fully achieve their stated security goals and should not be used alone in high-risk evidentiary contexts.Evidence B · P0
  15. MWITA-CD-2026-036YouTube requires creators to disclose realistic content that has been meaningfully altered or synthetically generated, including misleading depictions of real people or events.Evidence B · P1
  16. MWITA-CD-2026-037YouTube’s “Captured with a camera” disclosure requires C2PA 2.1 or later and acknowledges that broken provenance chains and air-gapped recapture can prevent the signal.Evidence B · P1
  17. MWITA-CD-2026-054Sony presented XYN at CES 2025 as a spatial-content creation workflow intended to capture and reproduce objects, human motion and backgrounds for use in 3D computer-graphics production.Evidence C · P2
  18. MWITA-CD-2026-057Google announced Veo 3 in May 2025 with generation of video plus ambient sound and dialogue, and made it available initially through Gemini and Flow in the United States and Vertex AI for enterprise users.Evidence B · P1
  19. MWITA-CD-2026-058Google expanded Veo 3 access to more than 150 countries and introduced eight-second Gemini photo-to-video clips with sound in July 2025, stating that generated videos carry both a visible watermark and invisible SynthID watermark.Evidence B · P1
  20. MWITA-CD-2026-059Adobe's April 2025 Firefly release moved its video model out of beta, supported clips up to five seconds at 1080p, introduced partner models in Firefly and stated that generated outputs receive Content Credentials identifying whether Firefly or a partner model was used.Evidence B · P1
  21. MWITA-CD-2026-060ByteDance published the Seedance 1.0 technical report in June 2025 describing text- and image-to-video generation with native multi-shot storytelling, up to 1080p and ten-second outputs, plus vendor-reported evaluation results.Evidence C · P2
  22. MWITA-CD-2026-061Alibaba announced in April 2025 that Wan2.1-FLF2V-14B was available through GitHub, Hugging Face and ModelScope for first-and-last-frame-conditioned five-second 720p video generation.Evidence B · P2
  23. MWITA-CD-2026-062Tencent's Hunyuan3D 2 repository released a two-stage image-conditioned pipeline for mesh generation followed by texture synthesis, with downloadable code and model assets for local 3D workflows.Evidence B · P2
  24. MWITA-CD-2026-063ByteDance announced Seed Audio 1.0 in July 2026 as a BytePlus service generating coordinated speech, sound effects and ambience, with voice conditioning from text or an authorized reference sample and outputs up to two minutes per pass.Evidence C · P2
  25. MWITA-CD-2026-064Google documents SynthID as embedding detectable watermarks in Google-generated images, video, audio and text, with Gemini able to check uploaded image, video or audio for a SynthID signal.Evidence B · P1
  26. MWITA-CD-2026-065Adobe launched the Content Authenticity app in public beta in April 2025, allowing creators to batch-apply Content Credentials to up to 50 JPG or PNG files and signal a preference against generative-AI training and use.Evidence B · P1
  27. MWITA-CD-2026-066Deepfake-Eval-2024 reports that evaluated open-source state-of-the-art detectors lost 50% video AUC, 48% audio AUC and 45% image AUC relative to older benchmarks on a 52-language in-the-wild dataset collected from 88 websites; commercial systems remained below forensic analysts.Evidence B · P0
  28. MWITA-CD-2026-067A NeurIPS 2025 Datasets and Benchmarks paper evaluating 15 detectors on 12,832 recaptured videos found Moiré artifacts degraded performance by as much as 25.4%, synthetic Moiré reduced accuracy by 21.4%, and tested demoiréing methods worsened accuracy by up to 16%.Evidence A · P0
  29. MWITA-CD-2026-068The UK government's March 2026 copyright-and-AI report found insufficient evidence of positive economic effect from the existing computer-generated-works protection and proposed its removal, while separately identifying gaps around unauthorized digital replicas and proposing further exploration rather than an enacted new personality right.Evidence A · P0
  30. MWITA-CD-2026-069YouTube stated in March 2026 that likeness detection launched to Partner Program creators in 2025 and expanded as a pilot to selected officials, journalists and political candidates, while explicitly noting that a match does not guarantee removal.Evidence B · P1
  31. MWITA-CD-2026-070YouTube announced in February 2026 that auto dubbing was available to all channels in 27 languages and reported more than six million daily viewers in December who watched at least ten minutes of auto-dubbed content, while retaining original-language preferences and creator controls.Evidence B · P1
  32. MWITA-AP-2026-003China's National Energy Administration reported more than 430 GW of new wind and solar generating capacity in 2025.Evidence A · P1
  33. MWITA-AP-2026-005India's MNRE records 28,723.65 MW of renewable-capacity additions excluding large hydro in FY2024-25, including 23,832.87 MW solar and 4,151.31 MW wind.Evidence A · P1
  34. MWITA-AP-2026-009Indonesia's energy ministry reported renewable sources at 15.75% of the 2025 electricity mix and 15,630 MW installed, led by hydro, bioenergy and geothermal rather than solar and wind.Evidence B · P1
  35. MWITA-AP-2026-023Singapore's first Biennial Transparency Report projected 2025 energy-sector emissions of 52.60 MtCO2e under its with-measures scenario, 7.20% above 2022.Evidence A · P1
  36. MWITA-AP-2026-027The Pacific Islands Forum's 2025 progress report identifies Digital Earth Pacific as a regional platform using decades of Earth-observation data within Blue Pacific collective action.Evidence B · P1
  37. MWITA-LC-2026-004Mexico's INDAUTOR stated that the Supreme Court decision it recognized in August 2025 determined that works created by artificial intelligence cannot be registered for copyright in Mexico.Evidence A · P1
  38. MWITA-ME-2026-013UAE Media Council Decision 3 of 2025 introduced an Advertiser Permit applying to individuals who produce advertising content from within the UAE on social media or modern technological means, with or without compensation.Evidence A · P0
  39. MWITA-ME-2026-014Jordan's Media Commission says Digital Media Regulation No. 13 of 2026 requires licensing for professional income-generating production or advertising creators, excludes personal non-professional accounts, and requires clear disclosure of AI use in regulated media content; the regulation took effect 23 May 2026.Evidence A · P0
  40. MWITA-RD-2026-007China's Measures for Labeling AI-Generated and Synthetic Content took effect on 1 September 2025, requiring covered providers to use explicit and metadata-based implicit labels and requiring dissemination services to inspect metadata and add notices in specified cases.Evidence A · P1
  41. MWITA-RD-2026-008China's GB 45438-2025, Cybersecurity Technology—Labeling Method for Content Generated by Artificial Intelligence, was published on 28 February 2025 and became effective on 1 September 2025 as a current national standard defining explicit and implicit labeling methods.Evidence A · P1
  42. MWITA-RD-2026-018C2PA Technical Specification 2.2 was published in May 2025, adding features including soft-binding manifest recovery, multipart assets, updated timestamp/revocation handling and a dedicated claim-signing extended key usage.Evidence A · P1
  43. MWITA-RD-2026-019The U.S. Copyright Office's January 2025 policy report concluded that generative-AI outputs are copyrightable only where a human author determines sufficient expressive elements; prompts alone do not provide sufficient control under the Office's analysis.Evidence A · P1
  44. MWITA-WD-2026-001The European Commission published a mandatory-format template on 24 July 2025 establishing a common minimum baseline for public summaries of content used to train general-purpose AI models under the AI Act.Evidence A · P0
  45. MWITA-WD-2026-002The European Commission published the General-Purpose AI Code of Practice on 10 July 2025 and endorsed it as an adequate voluntary route for signatories to demonstrate compliance with relevant AI Act transparency, copyright, safety and security obligations.Evidence A · P0
  46. MWITA-WD-2026-005From 19 May 2026, the US TAKE IT DOWN Act requires covered platforms to provide a notice-and-removal process and remove validly reported nonconsensual intimate imagery, including AI-created digital forgeries, and known identical copies within 48 hours.Evidence A · P0
  47. MWITA-WD-2026-006The US Copyright Office's May 2025 pre-publication Part 3 report analyzes generative-AI training under existing copyright law and licensing markets; it is an expert agency policy analysis, not legislation or a final court ruling that all training is fair use or infringement.Evidence A · P0
  48. MWITA-WD-2026-014Canada's Digital Regulators Forum published a September 2025 cross-regulator analysis mapping synthetic-media risks across privacy, competition, communications, copyright and automated-decision frameworks without creating a new unified synthetic-media law.Evidence A · P1
  49. MWITA-WD-2026-022The ACCC's final Digital Platform Services Inquiry report, released in June 2025, identified emerging competition and consumer issues in generative AI alongside messaging, app stores, mobile operating systems, ad tech, marketplaces, cloud and gaming, and recommended continued monitoring and broader reform.Evidence A · P1
  50. MWITA-BM-2026-014The Content Credentials initiative states that the C2PA specification ecosystem involves more than 500 companies led by major technology, media and provenance organizations.Evidence B · P1
  51. MWITA-BM-2026-015Adobe launched the public beta of its free Content Authenticity app in April 2025, supporting batch application of Content Credentials to up to 50 JPG or PNG files and reporting more than 4,500 Content Authenticity Initiative members.Evidence B · P1
  52. MWITA-XR-2026-005In a 2025 evaluation with 40 blind or low-vision participants, VideoA11y descriptions were selected over novice human descriptions in 90% of 200 comparisons and received higher ratings for descriptiveness, objectivity, accuracy and clarity.Evidence A · P1
  53. MWITA-XR-2026-006A 2025 mixed-methods study with 168 Deaf or hard-of-hearing survey participants, 15 interviewees and five professional captioners found persistent challenges in representing non-speech information and emphasized user agency and broader community involvement.Evidence A · P1
  54. MWITA-XR-2026-007A 2025 CHI study estimated material exclusion of disabled and older users across nine consumer VR/AR experiences; controller use alone was estimated to exclude about 5% of the UK population.Evidence B · P1
  55. MWITA-XR-2026-008W3C's XR Accessibility User Requirements identify multimodal support, input/output synchronization and customization as cross-disability needs for immersive environments.Evidence A · P1
  56. MWITA-XR-2026-009In a 2025 randomized study of 62 Chinese viewers, integrated subtitles placed near speakers did not improve comprehension versus standard subtitles, increased self-reported cognitive load and were perceived less favorably.Evidence A · P1
  57. MWITA-XR-2026-010A 2026 South African student study found participants reported subtitles enhanced understanding and engagement more than dubbing across academic English, plain English and home-language versions.Evidence B · P1
  58. MWITA-XR-2026-011The IWSLT 2025 evaluation campaign included automatic subtitling/dubbing, speech-to-speech translation, dialect, low-resource and Indic-language tasks with 32 participating teams.Evidence A · P1
  59. MWITA-XR-2026-012The 2025 VideoVista-CulturalLingo benchmark found 24 video-language models performed worse on Chinese-centric than Western-centric questions, especially Chinese history, and open models reached at most 45.2% on event localization.Evidence A · P1
  60. MWITA-XR-2026-015A 2025 randomized experiment with 911 participants found synthetic-media labels increased belief that images were AI-generated but did not significantly change intended liking, commenting or sharing; label design materially affected trust.Evidence A · P1
  61. MWITA-XR-2026-016C2PA Specification 2.2 defines cryptographically verifiable manifests, hard/soft bindings, update manifests, timestamps and revocation information for media provenance.Evidence A · P1
  62. MWITA-FC-2026-001IEA estimates global data-centre investment nearly doubled since 2022 to about USD 500 billion in 2024, while data centres consumed about 415 TWh or 1.5% of global electricity.Evidence A · P1
  63. MWITA-FC-2026-002IEA reports five large technology companies spent more than USD 400 billion in capex in 2025 and expected a further 75% increase in 2026, while warning that not all announced projects will come to fruition.Evidence A · P1
  64. MWITA-FC-2026-004IEA finds grid connections, generation, transformers, turbines, advanced chips and approvals are limiting near-term data-centre buildout despite booming investment pipelines.Evidence A · P1
  65. MWITA-FC-2026-019The Baku-to-Belém Roadmap calls for scaling all-source climate finance to developing countries to at least USD 1.3 trillion per year by 2035; it is a coordination roadmap, not money already committed or disbursed.Evidence A · P1
  66. MWITA-FC-2026-020OECD reports developed countries provided and mobilized USD 136.7 billion for developing-country climate action in 2024, above the USD 100 billion goal but far below the later USD 1.3 trillion roadmap aspiration.Evidence A · P1
  67. MWITA-FC-2026-021World Bank reports IBRD/IDA climate finance with co-benefits rose from USD 31.0 billion in FY2024 to USD 39.2 billion in FY2025.Evidence A · P1
  68. MWITA-ST-2026-001The FBI attributed the theft of approximately USD 1.5 billion in virtual assets from Bybit on or about 21 February 2025 to DPRK TraderTraitor actors, who dispersed converted assets across thousands of blockchain addresses.Evidence A · P1
  69. MWITA-ST-2026-002Coinbase reported that a threat actor paid multiple overseas support contractors or employees to collect customer and internal data from systems they were authorized to access for work; passwords, private keys and direct access to customer funds were not compromised.Evidence A · P1
  70. MWITA-ST-2026-003Coinbase preliminarily estimated USD 180–400 million of remediation and voluntary reimbursement expense from the incident, while explicitly stating that the full financial impact was unknown and the estimate could change materially.Evidence A · P1
  71. MWITA-ST-2026-006Marks & Spencer reported reduced product availability, manual-process waste and logistics costs, and a pause in online shopping; it estimated an approximately GBP 300 million 2025/26 operating-profit impact before mitigation, insurance and trading actions.Evidence A · P1
  72. MWITA-ST-2026-013The FBI's IC3 received 1,008,597 complaints in 2025 with USD 20.877 billion in reported losses, a 26% loss increase from 2024.Evidence A · P1
  73. MWITA-ST-2026-014The FBI reported that Operation Level Up had notified more than 8,000 cryptocurrency-investment-fraud victims and reduced losses by more than USD 500 million since inception.Evidence A · P1
  74. MWITA-ST-2026-015FTC Consumer Sentinel data recorded more than USD 12.5 billion in reported fraud losses for 2024, up 25%, while fraud reports remained about 2.6 million; the share of reporters saying they lost money rose from 27% to 38%.Evidence A · P1
  75. MWITA-ST-2026-016Singapore Police reported 37,308 scam cases in 2025, down 27.6%, and SGD 913.1 million lost, down 17.9%; 81.8% of reported scam cases involved victims making the transfers themselves.Evidence A · P1
  76. MWITA-ST-2026-017Singapore Police reported recovering more than SGD 139 million in scam proceeds and avoiding at least SGD 348 million in potential losses through alerts and interventions in 2025.Evidence A · P1
  77. MWITA-ST-2026-018UK Finance's 2025 Annual Fraud Report recorded more than GBP 1.1 billion of payment-fraud losses in 2024, with authorized push-payment fraud falling while remote-purchase fraud rose.Evidence B · P1
  78. MWITA-ST-2026-020In a synthetic retail-payments experiment, BIS Project Hertha found that combining payment-system analytics with bank or PSP analytics identified 12% more illicit accounts overall and improved results for new typologies by 26% relative to bank/PSP analytics alone.Evidence A · P1
  79. MWITA-ST-2026-021The Financial Stability Board finalized FIRE in April 2025 as a flexible common format and machine-readable taxonomy for financial cyber and operational incident reporting across jurisdictions.Evidence A · P1
  80. MWITA-CT-2026-032Singapore reported Design for Manufacturing and Assembly adoption in new development gross floor area rose from about 61% in 2023 to 68% in 2024.Evidence A · P1
  81. MWITA-MFG-2026-001UNIDO estimated global manufacturing production increased 0.50% quarter on quarter in Q4 2025.Evidence A · P0
  82. MWITA-MFG-2026-002UNIDO reported China led regional manufacturing growth in Q4 2025, followed by the rest of Asia-Pacific, while output elsewhere stagnated or declined and Africa recorded moderate growth.Evidence A · P0
  83. MWITA-MFG-2026-003UNIDO reported Q4 2025 output in lower-technology manufacturing rose only 0.2%, while medium-high and high-technology industries sustained quarterly growth above 1%. Evidence A · P1
  84. MWITA-MFG-2026-004The 2025 GVC Development Report put global-value-chain trade at 46.3% of world trade in 2024, down only modestly from a 48% peak in 2022.Evidence A · P0
  85. MWITA-MFG-2026-005The GVC Development Report 2025 concluded recent rewiring mostly benefited economies already established as suppliers rather than automatically deconcentrating production.Evidence A · P0
  86. MWITA-MFG-2026-006IEA estimated that excluding rare earths, the average share of the top refining country rose from 70% in 2023 to 72% in 2025.Evidence A · P0
  87. MWITA-MFG-2026-007IEA estimated full implementation of 2025 rare-earth export controls could put US$6.5 trillion per year of downstream production outside China at risk.Evidence A · P0
  88. MWITA-MFG-2026-008IEA estimated critical-mineral mining investment fell 9% in 2025, with battery-metals spending down more than 20% and lithium-company investment down about 40%.Evidence A · P0
  89. MWITA-MFG-2026-009IEA estimated advanced-economy public-finance commitments for critical minerals reached about US$65 billion in 2025, more than four times 2023, while warning commitments and disbursements differ.Evidence A · P1
  90. MWITA-MFG-2026-010IEA estimated Latin America and the Caribbean produces around one-quarter of global lithium but refines only about one-fifth of its mined output of key energy minerals other than lithium.Evidence A · P0
  91. MWITA-MFG-2026-011IFR counted 542,000 industrial robots installed worldwide in 2024, the fourth consecutive year above 500,000.Evidence A · P0
  92. MWITA-MFG-2026-012IFR counted 295,000 industrial robots installed in China in 2024, 54% of global deployments, with operational stock above two million.Evidence A · P0
  93. MWITA-MFG-2026-013IFR counted a record 9,100 industrial robot installations in India in 2024, 7% more than in 2023.Evidence A · P1
  94. MWITA-MFG-2026-014IFR counted 30,600 industrial robot installations in Korea in 2024, down 3%, and described annual installations as roughly flat near 31,000 since 2019.Evidence A · P1
  95. MWITA-MFG-2026-015China's statistics office reported production of 773,074 industrial robots in 2025, up 28.0% year on year.Evidence A · P0
  96. MWITA-MFG-2026-016China's statistics office reported 484.3 billion integrated circuits produced in 2025, up 10.9% year on year.Evidence A · P0
  97. MWITA-MFG-2026-017China's official communiqué reported manufacturing value added grew 6.4% in 2025.Evidence A · P1
  98. MWITA-MFG-2026-018India reported in April 2026 that commercial production had commenced at two of ten approved semiconductor projects, with two more expected to start during 2026.Evidence A · P0
  99. MWITA-MFG-2026-019India reported electronics-goods production of about ₹12 trillion in 2024/25, roughly six times the nominal value in 2014/15.Evidence A · P0
  100. MWITA-MFG-2026-020India reported 38 Electronics Components Manufacturing Scheme plants operational and 16 in advanced construction by 17 August 2026.Evidence A · P0
  101. MWITA-MFG-2026-021India's 106 approved ECMS projects carried projected investment of ₹695.48 billion, projected production of ₹5.341 trillion and expected direct employment of 74,628.Evidence A · P1
  102. MWITA-MFG-2026-022India approved the National Critical Mineral Mission with ₹163 billion government expenditure and ₹180 billion expected PSU investment over seven years.Evidence A · P1
  103. MWITA-MFG-2026-023India reported 58 entities pledging 850,000 tonnes per year of critical-mineral recycling capacity under its incentive scheme, versus a 270,000-tonne target.Evidence A · P0
  104. MWITA-MFG-2026-024Indonesia broke ground in June 2025 on an integrated battery project designed for 6.9 GWh of cell capacity with a stated later expansion to 15 GWh.Evidence A · P0
  105. MWITA-MFG-2026-025Indonesia's government described HLI Green Power's Karawang facility as operating with 10 GWh annual nameplate capacity, equivalent by design to batteries for about 150,000 EVs.Evidence A · P0
  106. MWITA-MFG-2026-026Malaysia's statistics office reported manufacturing sales of RM1.97 trillion in 2025, up 4.2% in nominal value from 2024.Evidence A · P1
  107. MWITA-MFG-2026-027INEGI recorded 3,953,494 light vehicles produced in Mexico in 2025, 0.9% fewer than in 2024.Evidence A · P0
  108. MWITA-MFG-2026-028Statistics South Africa reported total manufacturing production fell 1.3% in 2025 from 2024.Evidence A · P0
  109. MWITA-MFG-2026-029Morocco's 2025 industry survey reported 2024 industrial production value of MAD842 billion, up 12%, and industrial employment of 1,038,133, a net increase of 42,714 jobs.Evidence A · P1
  110. MWITA-MFG-2026-030Saudi Arabia's preliminary industrial index showed manufacturing output 6.3% higher year on year in September 2025.Evidence A · P1
  111. MWITA-MFG-2026-031Brazil's November 2025 tax-advantaged debenture policy estimated potential annual investment of R$5.2 billion in strategic-mineral projects, including R$3.7 billion for transformation and R$1.5 billion for mining and beneficiation.Evidence A · P1
  112. MWITA-MFG-2026-032Brazil's 2026 official investor guide says domestic production of high-purity inputs for energy-transition value chains remains incipient despite substantial critical-mineral potential.Evidence A · P0
  113. MWITA-CE-2026-004ITU estimated 5G covered 55% of the world population in 2025, ranging from 84% in high-income economies to 4% in low-income economies.Evidence A · P0
  114. MWITA-CE-2026-007ITU reported data centres account for about 1.5% of global electricity consumption.Evidence A · P1
  115. MWITA-CE-2026-008China's MIIT reported 4.838 million 5G base stations at end-2025, up 588,000 during the year.Evidence A · P0
  116. MWITA-CE-2026-009MIIT reported 2.064 million 5G base stations supported RedCap at end-2025, 42.7% of 5G stations.Evidence A · P1
  117. MWITA-CE-2026-010China's three state telecom operators reported 938,000 externally serving data-centre racks at end-2025, 108,000 more than a year earlier.Evidence A · P1
  118. MWITA-CE-2026-031MIDA reported actual data-centre electrical load utilization reached 710 MW in September 2025, up from 405 MW at end-2024.Evidence A · P0
  119. MWITA-CE-2026-032By September 2025, 49 Malaysian data centres had signed electricity-supply agreements representing 7.1 GW maximum demand, to be utilized in phases over five to ten years.Evidence A · P0
  120. MWITA-CE-2026-033Singapore's second data-centre call required proposed facilities to target PUE of 1.25 or better at full IT load and at least 50% eligible green energy pathways.Evidence A · P0
  121. MWITA-DA-2026-035Google described Expressive Captions in 2025 as real-time captions across most Android apps that aim to convey how speech is delivered as well as its words.Evidence B · P1
  122. MWITA-CA-2026-001C2PA published Content Credentials specification version 2.2 in May 2025.Evidence A · P1
  123. MWITA-CA-2026-002C2PA 2.2 binds assertions, a signed claim and its claim signature into a verifiable Manifest associated with an asset.Evidence A · P1
  124. MWITA-CA-2026-003C2PA 2.2 explicitly avoids judging provenance assertions as good or bad and instead validates association, formation and tamper state.Evidence A · P1
  125. MWITA-CA-2026-004C2PA’s 2.2 explainer describes durable Content Credentials as combining cryptographic hard binding with soft binding such as watermarking or fingerprinting.Evidence A · P1
  126. MWITA-CA-2026-005Under C2PA 2.2, a trusted Manifest requires a valid signature whose signing certificate chains to a trust anchor accepted by the validator.Evidence A · P1
  127. MWITA-CA-2026-006C2PA launched its Conformance Program and official Trust List in mid-2025.Evidence A · P1
  128. MWITA-CA-2026-007C2PA conformance provides assurance that a tested product produces or validates C2PA data correctly against defined requirements.Evidence A · P1
  129. MWITA-CA-2026-008China’s Measures for Labeling AI-Generated and Synthetic Content took effect on 1 September 2025.Evidence A · P1
  130. MWITA-CA-2026-009China’s 2025 measures require perceptible labels for covered AI-generated text, images, audio, video and virtual scenes, with modality-specific presentation rules.Evidence A · P1
  131. MWITA-CA-2026-010China’s 2025 measures require covered providers to add implicit metadata containing synthetic-content attributes and production information such as provider and content identifiers.Evidence A · P1
  132. MWITA-CA-2026-011China’s measures require distribution services to inspect metadata, use user declarations and detection signals, and apply differentiated labels for generated, possibly generated or suspected generated content.Evidence A · P1
  133. MWITA-CA-2026-012China’s 2025 measures prohibit malicious deletion, alteration, forgery or concealment of required synthetic-content labels and prohibit providing tools or services for those acts.Evidence A · P1
  134. MWITA-CA-2026-013China published mandatory national standard GB 45438-2025 for AI-generated and synthetic-content labeling on 28 February 2025, effective 1 September 2025.Evidence A · P1
  135. MWITA-CA-2026-014GB 45438-2025 supplies the labeling method while the companion 2025 Measures allocate legal duties among service and distribution actors.Evidence A · P1
  136. MWITA-CA-2026-015South Korea’s AI Basic Act entered into full force on 22 January 2026.Evidence A · P1
  137. MWITA-CA-2026-016South Korea’s official 2026 guidance says downloaded or shared AI-generated text, images or videos must be labeled either with human-recognizable visible/audible identifiers or machine-readable metadata after notice.Evidence A · P1
  138. MWITA-CA-2026-017South Korea’s labeling duty identifies AI generation or manipulation status; it does not certify that labeled content is false or that unlabeled content is authentic.Evidence A · P1
  139. MWITA-CA-2026-018India notified amendments concerning synthetically generated information through Gazette notification G.S.R. 120(E) on 10 February 2026 and issued corrigenda on 26 February 2026.Evidence A · P1
  140. MWITA-CA-2026-019India’s 2026 SGI amendments require prominent labeling of covered permissible synthetic information and permanent metadata or other technical provenance mechanisms to the extent technically feasible.Evidence A · P1
  141. MWITA-CA-2026-020India’s 2026 framework adds due-diligence duties for significant social-media intermediaries involving user declarations, reasonable technical verification and prominent SGI labels.Evidence A · P1
  142. MWITA-CA-2026-021The EU AI Act’s Article 50 transparency obligations apply from 2 August 2026.Evidence A · P1
  143. MWITA-CA-2026-022Article 50 requires providers of AI systems generating synthetic audio, image, video or text to make outputs machine-readable and detectable as artificially generated or manipulated, subject to the law’s conditions.Evidence A · P1
  144. MWITA-CA-2026-023Article 50 requires deployers to disclose AI-generated or manipulated image, audio or video that constitutes a deepfake, with tailored treatment for evidently artistic, creative, satirical or fictional works.Evidence A · P1
  145. MWITA-CA-2026-024The EU transparency Code of Practice is voluntary, while the underlying Article 50 marking and disclosure duties are legal obligations.Evidence A · P1
  146. MWITA-CA-2026-025Google announced the SynthID Detector portal on 20 May 2025 and initially rolled it out to early testers.Evidence C · P1
  147. MWITA-CA-2026-026At launch, SynthID Detector was designed to identify SynthID watermarks in content made with Google AI across image, audio, video and text modalities.Evidence C · P1
  148. MWITA-CA-2026-027Google stated in May 2025 that more than 10 billion pieces of content had been watermarked with SynthID.Evidence C · P1
  149. MWITA-CA-2026-028Google disclosed in February 2025 that some small Reimagine edits may be too limited for SynthID to label and detect.Evidence C · P1
  150. MWITA-CA-2026-029TikTok reported in September 2025 that it had expanded C2PA Content Credentials to additional features and added a visible watermark for AI content created with the TikTok camera.Evidence C · P1
  151. MWITA-CA-2026-030TikTok reported that zero-view removals for content violating its Civic and Election Integrity rules increased by 15 percentage points to 90% in its sixth Disinformation Code report.Evidence C · P1
  152. MWITA-CA-2026-031TikTok reported banning nearly 3,000 government, politician and political-party impersonation accounts in the period covered by its 2025 report.Evidence C · P1
  153. MWITA-CA-2026-032RobustSora introduced a 6,500-video benchmark with authentic-clean, authentic-with-fake-watermark, generated-watermarked and generated-de-watermarked classes.Evidence B · P1
  154. MWITA-CA-2026-033Across ten tested models, RobustSora reported performance changes of 2–8 percentage points under watermark manipulation.Evidence B · P1
  155. MWITA-CA-2026-034RobustSora explicitly evaluates whether adding fake watermark cues to authentic videos triggers detector false alarms.Evidence B · P1
  156. MWITA-CA-2026-035VoiceWukong’s February 2025 dataset release includes 265,200 English and 148,200 Chinese deepfake-voice samples.Evidence B · P1
  157. MWITA-CA-2026-036VoiceWukong was designed to compare humans, dedicated detectors and a multimodal language model in voice-deepfake detection scenarios.Evidence B · P1
  158. MWITA-CA-2026-037SpeechFake, published at ACL 2025, was built as a large-scale multilingual speech-deepfake dataset covering varied generation methods, languages and speakers.Evidence A · P1
  159. MWITA-CA-2026-038SpeechFake’s experiments test how generation method, language diversity and speaker variation affect detector performance rather than assuming one aggregate score transfers across conditions.Evidence A · P1
  160. MWITA-CA-2026-039The U.S. Copyright Office published Part 2 of its Copyright and Artificial Intelligence report on 29 January 2025, addressing copyrightability of generative-AI outputs.Evidence A · P1
  161. MWITA-CA-2026-040The U.S. Copyright Office released a pre-publication Part 3 report on generative-AI training on 9 May 2025 and stated that a final version would follow without expected substantive changes.Evidence A · P1
  162. MWITA-CA-2026-041EU obligations applying to general-purpose AI model providers from 2 August 2025 include implementing a policy to comply with Union copyright law.Evidence A · P1
  163. MWITA-CA-2026-042EU GPAI obligations applying from 2 August 2025 require providers to publish a sufficiently detailed summary of content used to train the model.Evidence A · P1
  164. MWITA-CA-2026-043Korea’s MCST released an English Guide to Copyright Registration for Generative AI-Assisted Works on 3 November 2025 after its Korean release in June.Evidence A · P1
  165. MWITA-CA-2026-044Korea’s companion 2025 guide identifies legal principles and factors for assessing whether a generative-AI output infringes an existing work.Evidence A · P1
  166. MWITA-CA-2026-045MCST frames its dispute-prevention guidance separately for rights holders, users and AI service providers.Evidence A · P1
  167. MWITA-CA-2026-046Japan’s Agency for Cultural Affairs says autonomously generated output is generally not a work, while output made with AI as a tool may qualify when a person has creative intent and makes a creative contribution.Evidence A · P1
  168. MWITA-CA-2026-047The Japanese General Understanding on AI and Copyright is expressly non-binding and not a definitive legal assessment of a specific generative-AI system.Evidence A · P1
  169. MWITA-WS-2026-010DEWA reported 495 million imperial gallons per day of installed desalination capacity in 2025.Evidence A · P1
  170. MWITA-WS-2026-011DEWA reported desalinated-water production of 158,359 million imperial gallons in 2025.Evidence A · P1
  171. MWITA-WS-2026-012DEWA reported average daily water demand of 442.478 million imperial gallons and a peak monthly average of 477 million imperial gallons in 2025.Evidence A · P1
  172. MWITA-WS-2026-014DEWA describes the Hassyan seawater reverse-osmosis project as planned to supply up to 180 million imperial gallons per day from the first quarter of 2027.Evidence A · P1
  173. MWITA-WS-2026-015DEWA states an investment value of AED 3.377 billion for the Hassyan desalination project.Evidence A · P1
  174. MWITA-WS-2026-016Singapore PUB says the Deep Tunnel Sewerage System and Tuas Water Reclamation Plant are scheduled for phased commissioning from 2027.Evidence A · P1
  175. MWITA-WS-2026-017PUB expects the Tuas Water Reclamation Plant to use about half the land footprint of conventional facilities and free roughly 150 hectares elsewhere.Evidence A · P1
  176. MWITA-WS-2026-020India reported household tap-water connections for 15.76 crore of 19.36 crore rural households, or 81.42 percent, in December 2025.Evidence A · P1
  177. MWITA-WS-2026-021India reported an increase of 12.53 crore rural household tap-water connections from the 2019 baseline to December 2025.Evidence A · P1
  178. MWITA-WS-2026-022India states a central Jal Jeevan Mission outlay of ₹2,08,652 crore for the programme period cited in its December 2025 release.Evidence A · P1
  179. MWITA-WS-2026-028Brazil amended its National Water Resources Policy in 2025 to add water and energy security through storage among policy objectives.Evidence A · P1
  180. MWITA-WS-2026-030A World Bank results framework records public consultation on Morocco's draft desalination decree as a 2025 programme milestone.Evidence A · P1
  181. MWITA-WS-2026-031The Morocco programme framework schedules adoption of a desalination decree and reuse-related implementing rules for 2026.Evidence A · P1
  182. MWITA-WS-2026-032The Mohamed bin Zayed Water Initiative and World Bank signed a 2025 memorandum to explore water innovation and investment collaboration.Evidence A · P1
  183. MWITA-WS-2026-034Australia announced A$22.6 million for non-urban water-meter telemetry and anticipated more than 5,000 new telemetry devices.Evidence A · P1
  184. MWITA-II-2026-027ARC presented just over USD 5.4 million in parametric payouts to Mozambique and the World Food Programme in October 2025 for drought and Cyclone Chido response.Evidence A · P1
  185. MWITA-II-2026-028The Mozambique payout comprised just over USD 1.8 million to government for drought response and USD 3.5 million to WFP across drought and cyclone policies.Evidence A · P1
  186. MWITA-II-2026-029WFP's Cyclone Chido policy component released USD 3.1 million for response in Nampula and Cabo Delgado.Evidence A · P1
  187. MWITA-II-2026-030Malawi received USD 3,376,783 from ARC in July 2025 after drought in the 2024/25 agricultural season.Evidence A · P1
  188. MWITA-II-2026-031Malawi's ARC total combined a USD 3,065,746 traditional sovereign-policy payout and a USD 311,037 anticipatory-policy payout.Evidence A · P1
  189. MWITA-II-2026-032Malawi planned to use the traditional payout for food and cash support to 311,000 beneficiaries and the anticipatory payout for crop inputs to 3,255 households.Evidence A · P1
  190. MWITA-II-2026-033Madagascar became the first subscriber to ARC's sovereign parametric flood policy when the contract was signed in January 2026.Evidence A · P1
  191. MWITA-II-2026-035ARC reports issuing more than 300 sovereign policies and paying over USD 250 million in claims since 2014.Evidence A · P1
  192. MWITA-II-2026-036South Africa's National Treasury reported in August 2025 that most public infrastructure was uninsured, creating a government contingent liability.Evidence A · P1
  193. MWITA-II-2026-037South Africa reported that municipal insurance-pool coverage was limited partly by poor asset data and maintenance records.Evidence A · P1
  194. MWITA-II-2026-038South Africa's Treasury and municipalities planned a parametric-insurance pricing and structure pilot by the end of the third quarter of 2025.Evidence A · P1
  195. MWITA-II-2026-042Saudi Arabia's Insurance Authority added 15 digital regulatory services to TAMEEN in December 2025 after launching the portal with 36 services in February.Evidence A · P1
  196. MWITA-LV-2026-028ONDC Sahayak launched as a WhatsApp bot in five languages, with an announced plan to extend to 22 languages.Evidence A · P1
  197. MWITA-UG-2026-002Tencent reported expanding game-engine compatibility for Weixin Mini Games in Q2 2025.Evidence B · P1
  198. MWITA-UG-2026-003Tencent reported reducing Weixin Mini Games loading time in Q2 2025.Evidence B · P1
  199. MWITA-UG-2026-019Fortnite engagement-payout eligibility requires creators to be at least 18 years old and able to receive Hyperwallet payments.Evidence A · P1
  200. MWITA-UG-2026-020Fortnite engagement-payout eligibility requires an account active for at least 30 days with editing activity on at least seven of those days.Evidence A · P1
  201. MWITA-UG-2026-036In Unity's 2025 developer research, 64% of surveyed developers said they were working on online multiplayer or couch co-op games.Evidence B · P1
  202. MWITA-UG-2026-037Unity reported that 90% of surveyed developers had launched a recent game on mobile and 80% on desktop.Evidence B · P1
  203. MWITA-UG-2026-038Unity reported that 96% of surveyed studios used AI tools in selected game-development workflows.Evidence B · P1
  204. MWITA-UG-2026-039Minecraft's 2025 summer update added an in-game music disc and painting tied to A Minecraft Movie.Evidence A · P1
  205. MWITA-UG-2026-040Microsoft reported Minecraft community challenges reaching 50 million happy-ghast completions and 1 million ghastling completions.Evidence B · P1
  206. MWITA-CMON-2026-023More than 1,200 people participated in note's 2025 pilot for compensated use of creator data in AI training.Evidence B · P1
  207. MWITA-CMON-2026-024note said it would distribute more than JPY5 million in total to creators whose data was used in its AI-training pilot.Evidence B · P1
  208. MWITA-CMON-2026-025note reported that the maximum individual distribution in its AI-training pilot would exceed JPY400,000.Evidence B · P1
  209. MWITA-SR-2026-035Kantar states LINK AI predicts ad-effectiveness scores in as little as 15 minutes without a consumer sample.Evidence C · P0
  210. MWITA-SR-2026-037The documented LINK AI pipeline decomposes media and extracts up to 20,000 features before predicting creative-effectiveness metrics.Evidence C · P0
  211. MWITA-SR-2026-070Automated creative scores should triage what receives scarce human testing, not silently replace market-specific validation for high-spend or high-risk campaigns.Evidence C · P0
  212. MWITA-SPC-2026-001China's 2023–2025 metaverse action plan calls for integrated advances across AI, blockchain, cloud computing and virtual reality, plus operating systems, middleware, modelling tools and rendering and simulation engines.Evidence A · P1
  213. MWITA-CO-2026-001Across 4,633 simultaneously launched sibling ads spanning more than 369 million impressions and 2.5 million clicks, AI-generated images had no detectable average CTR disadvantage versus human-made siblings; performance was better when AI images did not look AI-generated.Evidence A · P1
  214. MWITA-CO-2026-002Across laboratory and field studies, fully GenAI-created visual ads increased CTR by up to 19% versus human or GenAI-modified ads, while disclosure of AI involvement reduced effectiveness by up to 31.5%.Evidence B · P1
  215. MWITA-CO-2026-003In a fixed-video randomized online experiment, an AI label reduced selection probability by 23%, and an AI callout during viewing reduced watch time by 14% and willingness to engage with the creator's other videos.Evidence A · P1
  216. MWITA-CO-2026-004A randomized field experiment with 1,337 reached customers found that disclosure of an AI voice sales agent did not reduce conversion relative to nondisclosure.Evidence A · P1
  217. MWITA-CO-2026-005Three studies that separated true from perceived ad origin found AI- and human-generated ads broadly equivalent on most effectiveness measures, but ads believed human scored higher regardless of actual origin.Evidence B · P1
  218. MWITA-CO-2026-006A social-media field experiment and survey experiment found positive responses to disclosing human–ChatGPT collaboration; positive ChatGPT-only disclosure was significant only in the social-media field setting.Evidence B · P1
  219. MWITA-CO-2026-007In a paid-webcomic field experiment, a simple AI-generated-content disclosure reduced conversion, payment, viewed episodes and viewing time, while normative and technology framing largely restored metrics to nondisclosure levels; effort framing did not.Evidence B · P1
  220. MWITA-CO-2026-008Three laboratory experiments found that advertising disclosure attached to the product explicitly sought in a generative-AI answer reduced trust in both system and product, whereas disclosure on a later non-target recommendation reduced recommendation trust without a significant system-trust penalty.Evidence B · P1
  221. MWITA-CO-2026-009In a within-subject advertisement-design experiment, expert evaluators rated GenAI-supported work more creative and unconventional but found no significant visual-appeal, brand-alignment or usefulness gains; overall creative-thinking ability did not significantly improve.Evidence B · P1
  222. MWITA-CO-2026-010A multi-study program found higher processing fluency for generated versus conventional ads, but a humorous conventional ad retained higher brand recognition than its generated counterpart while two other categories showed near-identical recall and recognition.Evidence A · P1
  223. MWITA-CO-2026-011A field experiment with a Chinese electronics company found AI-powered targeting reduced click propensity but increased post-click purchase propensity relative to manual demographic targeting, with different patterns for popular, niche and innovative products.Evidence B · P1
  224. MWITA-CO-2026-012In a randomized online-travel-planning field experiment, enthusiastic GenAI elicited longer prompts, while enthusiastic and neutral variants were more likely than an unconstrained-tone control to generate web-service subscription purchases.Evidence B · P1
  225. MWITA-AIAO-2026-003A CTR-optimized advertising-image pipeline reported 7.4% higher click-through rate than a pretrained multimodal model in adaptive online serving.Evidence B · P1
  226. MWITA-AIAO-2026-005LLM-generated ads did not significantly outperform human ads on personality personalization but were preferred more often on general persuasion principles, with a large penalty when AI origin was correctly detected.Evidence B · P1
  227. MWITA-AIAO-2026-006A deceptively synthetic crowd image reduced multiple event-ad evaluations, while a disclaimer restored interest and stated attendance intention approximately to human-created-ad levels.Evidence A · P1
  228. MWITA-AIAO-2026-007AI-origin disclosure improved recognition in Instagram travel advertising but reduced brand credibility, ad attitude, and stated purchase intention for influencer-sourced content.Evidence A · P1
  229. MWITA-AIAO-2026-008In green advertising, disclosure that an image was AI-generated produced more favorable brand attitudes and purchase intentions than disclosure that the copy was AI-generated.Evidence A · P1
  230. MWITA-AIAO-2026-009Disclosing that both advertising copy and image were AI-produced reduced perceived advertising value and purchase intention, with credibility mediating the negative response.Evidence A · P1
  231. MWITA-AIAO-2026-010Consumer trust differed across five concealed in-app ad-generation modalities spanning AI, human, and user-generated advertisements and memes.Evidence A · P1
  232. MWITA-AIAO-2026-011Adding explanations or beneficiary framing to an AI disclosure did not repair advertising response in a preregistered replication; explanatory wording slightly worsened ad and brand attitudes while engagement intention was statistically equivalent.Evidence A · P1
  233. MWITA-AIAO-2026-012Product-relevant AIGC advertisements directed substantially more gaze toward the product region than divergent AIGC advertisements among Chinese university viewers.Evidence A · P1
  234. MWITA-SSCO-2026-001One generative image model outperformed professional stock imagery on clicks in one display-ad setting, while the other tested models did not.Evidence A · P1
  235. MWITA-SSCO-2026-002In Lebanese wine marketing, virtual influencers produced stronger engagement and purchase intention under a hedonic appeal, while human influencers were more persuasive under authenticity framing.Evidence A · P1
  236. MWITA-SSCO-2026-003Transparency labeling did not uniformly increase trust in a virtual influencer; its direction reversed with Korean participants’ machine-heuristic tendency.Evidence A · P1
  237. MWITA-SSCO-2026-004Human and virtual influencers activated different persuasion pathways in Spain, with the human advantage in parasocial relationship larger for hedonic than utilitarian products.Evidence A · P1
  238. MWITA-SSCO-2026-005Head-mounted viewing of identical 360-degree influencer advertising improved several immediate attitudes and intentions versus smartphone viewing, but purchase intention did not remain elevated after one month.Evidence A · P1
  239. MWITA-SSCO-2026-008People detected political speech deepfakes substantially better with audio or audiovisual information than from transcripts, but text-to-speech fakes remained far from perfectly recognized.Evidence A · P1
  240. MWITA-SSCO-2026-009In a large UK feed experiment, combining user-centered interventions with content labels improved truth judgments most; user interventions generally outperformed labels without significantly reducing trust in true posts.Evidence A · P1
  241. MWITA-SSCO-2026-010A Chinese sign-language-avatar benchmark showed measurable improvement but limited absolute quality: only four of ten submitted systems achieved good readability.Evidence B · P1
  242. MWITA-SSCO-2026-011Synthetic human presenters were not less effective than real presenters for identical business pitches on average; outcomes were stronger when audiences believed the presenter was human.Evidence A · P1
  243. MWITA-HCO-2026-001In a preregistered cervical-screening messaging experiment, AI-targeted and AI-tailored narratives were perceived as lower quality and less believable and did not increase screening motivation beyond nonnarrative or generic narratives.Evidence A · P1
  244. MWITA-AMPO-2026-006Human–AI collaboration improved creativity in a short ideation task, but did not outperform human–human collaboration and varied nonlinearly by baseline ability.Evidence A · P1
  245. MWITA-AMPO-2026-012Human–AI teams produced substantially more ads and better copy, but worse images, less diversity and no significant live-platform outcome lift.Evidence B · P1
  246. MWITA-ADTO-2026-001Labels on misleading AI images reduced belief and credibility, while a standalone AI-generated label did not significantly reduce sharing intention.Evidence A · P1
  247. MWITA-ADTO-2026-002An AI-generated label lowered perceived accuracy and sharing willingness even for true or human-made headlines.Evidence A · P1
  248. MWITA-ADTO-2026-003Specific transparency warnings improved deepfake recognition but did not eliminate effects on guilt and character judgments.Evidence A · P1
  249. MWITA-ADTO-2026-005People followed AI warning labels even when random or wrong, and false labels harmed news discrimination more than correct labels helped.Evidence A · P1
  250. MWITA-ADTO-2026-006Across five countries, AI-video warnings reduced perceived accuracy but did not improve true-versus-false discrimination.Evidence A · P1
  251. MWITA-ADTO-2026-007AI and misinformation labels lowered authenticity of marked images while increasing assumed authenticity of unmarked images.Evidence A · P1
  252. MWITA-ADTO-2026-008Disclosing fully automated generative-AI social content reduced post credibility, brand attitude and perceived authenticity; AI assistance produced a weaker penalty.Evidence A · P1
  253. MWITA-ADTO-2026-009Explaining AI's specific role in ad creation improved persuasion mainly through role clarity and process credibility, not a robust direct attitude lift.Evidence A · P1
  254. MWITA-ADTO-2026-010AI disclosure harmed trust and ad attitudes most when service advertising depicted intangible or human attributes, while tangible backgrounds largely avoided the penalty.Evidence A · P1
  255. MWITA-ADTO-2026-011Top teams in a large open competition removed tested invisible image watermarks from most images while retaining high visual quality.Evidence A · P1
  256. MWITA-ADTO-2026-012Political deepfakes harmed evaluations of targeted politicians even when viewers recognized them as less credible, and fact-checks did not reliably repair the damage.Evidence A · P1
  257. MWITA-ICO-2026-005AI disclosure did not reduce purchase intention overall, but influencer-sourced travel posts showed pairwise credibility and brand-attitude penalties.Evidence A · P1
  258. MWITA-ICO-2026-006A prominent retouching label did not help adolescents detect image manipulation, while undisclosed retouching produced the largest brand-attitude penalty.Evidence A · P1
  259. MWITA-ICO-2026-009For AI influencers, personal self-disclosure reduced eeriness and supported endorsement responses more reliably than photorealistic appearance alone.Evidence A · P1
  260. MWITA-RCO-2026-002Facial profile photos beside positive product reviews increased real purchase conversion, while non-facial profile personalization did not reproduce the effect.Evidence A · P1
  261. MWITA-RCO-2026-007The mediated association between eWOM susceptibility and purchase intention was stronger on third-party review platforms than on enterprise-owned review sites.Evidence C · P1
  262. MWITA-RCO-2026-009Affective brand engagement and virtual-brand-community engagement were distinct, positive correlates of brand loyalty and eWOM across four countries.Evidence C · P1
  263. MWITA-BSO-2026-004Google reported that permanently suspending more than 700,000 advertiser accounts associated with AI-generated public-figure impersonation scams coincided with a 90% decline in reports of that scam-ad type during 2024.Evidence B · P1
  264. MWITA-BSO-2026-012Spontaneously recognizing a deepfake user-generated-content-style advertisement produced more negative attitudes, engagement intentions and purchase intentions than either not recognizing it or recognizing it only after later disclosure.Evidence A · P1
  265. MWITA-ASDO-2026-011Adding social-proof language to otherwise controlled product descriptions consistently increased recommendation rate and rank across tested LLMs, whereas scarcity and exclusivity language reduced visibility.Evidence A · P1
  266. MWITA-CPCO-2026-001A pre-election mailing showing the household's and neighbors' past voting records increased validated primary-election turnout by 8.1 percentage points relative to the no-mail control; progressively weaker social-pressure treatments produced smaller effects.Evidence A · P1
  267. MWITA-CPCO-2026-002A Facebook Election Day message displaying friends who had clicked 'I Voted' increased validated voting by 0.39 percentage points versus no message (SEM 0.17; p=0.02), while the information-only message had a 0.00-point effect (SEM 0.28; p=0.98).Evidence A · P1
  268. MWITA-CPCO-2026-005Court-date text reminders reduced failure to appear by 8.0 percentage points from a 37.9% control rate (21% relative; p<0.001); the message combining logistical and consequence information reduced it by 9.9 points.Evidence A · P1
  269. MWITA-CPCO-2026-007In a randomized heavy-rain warning survey, the strongest peer-externality message increased stated shelter-evacuation intention by about 15 percentage points versus the control; at eight months, a different gain/prevention message increased reported emergency-item preparation by 4.7-5.8 points, but no message changed observed pre-evacuation.Evidence A · P1
  270. MWITA-CPCO-2026-009Prompting participants to rate the accuracy of one unrelated headline before evaluating COVID-19 headlines nearly tripled truth discernment in subsequent sharing intentions compared with control.Evidence A · P1
  271. MWITA-CPCO-2026-012Telling prospective donors that a lead donor had already covered overhead increased donation rate by 80% versus a seed-money appeal and 94% versus a 1:1 match; total donations rose 75% and 89%, respectively.Evidence A · P1
  272. MWITA-EBRO-2026-001Recruitment postcards emphasizing the challenge of policing or long-term career benefits generated about three times as many police applications as the control, whereas public-service-motivation messages did not attract people who would not otherwise have applied; the effective messages did not observably reduce applicant quality and were especially effective among women and people of color.Evidence A · P1
  273. MWITA-EBRO-2026-002Adding a conventional equal-employment-opportunity statement to real job advertisements reduced rather than increased racial-minority job seekers' likelihood of completing an application, with larger adverse responses among more educated job seekers and in white-majority cities.Evidence B · P1
  274. MWITA-EBRO-2026-003Gender-neutral wording increased the female share of applicants only when a job ad was saliently different from mostly untreated ads; the effect attenuated as applicants were exposed to more treated ads, implying that the small-scale diversity gain did not scale mechanically platform-wide.Evidence B · P1
  275. MWITA-EBRO-2026-004Telling directly sourced candidates that an AI application had been used to identify them significantly reduced their interest in a real public-sector job offer, even though an exploratory deployment analysis indicated that the tool's person-job-fit score could predict leaders' fit assessments.Evidence A · P1
  276. MWITA-EBRO-2026-007Advertising higher expected earnings attracted 36% more applicants and increased vacancy filling, but also signaled a less prosocial role and reduced recruitment of the most socially motivated candidates, who subsequently stayed longer and performed better.Evidence A · P1
  277. MWITA-EBRO-2026-009Performance-contingent referral rewards induced high-ability participants to refer higher-performing contacts, whereas larger fixed referral payments did not improve referral quality; high-ability referrers possessed useful private information but used it selectively when incentives aligned.Evidence A · P1
  278. MWITA-EBRO-2026-011Giving real applicants the interview questions immediately before a structured videoconference interview improved perceived chance to perform and overall fairness, but did not reduce anxiety; a standardized rapport-building script produced no significant applicant-reaction benefit.Evidence A · P1
  279. MWITA-EBRO-2026-012Negative employment word-of-mouth reduced applicant attraction most strongly for an unknown employer; established positive or negative employer-brand knowledge attenuated—but did not eliminate—the negative signal.Evidence A · P1
  280. MWITA-PCR-2026-001When the organization’s PR practitioner disclosed its own crisis before another source did, journalists rated the practitioner as more credible, but expressed greater—not lower—interest in the story; crisis-severity judgments and framing in reporters’ news leads did not differ.Evidence A · P1
  281. MWITA-PCR-2026-002Organizational self-disclosure before third-party revelation raised all three trust dimensions and purchase intention, but its advantage was conditioned by prior reputation and crisis type rather than being a universal timing effect.Evidence A · P1
  282. MWITA-PCR-2026-003Apology was not superior to compensation or sympathy: the three victim-centered responses produced statistically similar reputation, account-acceptance, anger and negative-word-of-mouth scores, while information-only drove the observed reputation and account-acceptance differences.Evidence A · P1
  283. MWITA-PCR-2026-004Crisis-communication channel affected reputation, secondary communication and boycott-related reactions, whereas message strategy affected only secondary crisis reactions; in this design the medium separated more outcomes than apology, sympathy or information content.Evidence A · P1
  284. MWITA-PCR-2026-005An apology repaired trust when hotel overcharging was attributed to an external computer failure, but not when the same 60% price increase was portrayed as intentional harm.Evidence A · P1
  285. MWITA-PCR-2026-006Passive wording reduced attributed responsibility and indirectly protected reputation only when the organization denied an unresolved accusation; it supplied no reputational benefit when paired with an apology.Evidence A · P1
  286. MWITA-PCR-2026-007Debunks from competitors, media or regulators substantially reversed misinformation-driven product avoidance in incentive-compatible choices; for fluoride toothpaste the share choosing the ingredient fell from .67 to .56 after misinformation and recovered to .64 after debunking.Evidence A · P1
  287. MWITA-PCR-2026-008Across five studies, fact-check labels reduced belief in repeated corporate misinformation without a familiarity backfire; strong unambiguous corrections could reverse the repetition effect, but brand evaluations were only partially restored.Evidence A · P1
  288. MWITA-PCR-2026-010Corporate-misconduct events had no average US stock reaction until at least four high-reach outlets covered them; at that threshold mean three-day abnormal return was -1.12%, revealing exposure scale as a market-value boundary.Evidence B · P1
  289. MWITA-PCR-2026-011For 169 Chinese product-harm crises, firm-first disclosure was associated with a more negative three-day market reaction than third-party disclosure (-1.74% versus -1.09% CAR), the opposite of the trust benefit often found in consumer experiments.Evidence B · P1
  290. MWITA-PCR-2026-012After a five-month peanut-butter recall, post-crisis advertising accelerated first repurchase for the stronger Kraft brand but not the weaker Eta brand; observed recovery depended on pre-crisis loyalty, familiarity and brand strength rather than communication alone.Evidence B · P1
  291. MWITA-ELO-2026-001In the main randomized field experiment, adding the recipient's name to the subject line raised detected opens from 9.05% to 10.80%, sales leads from 0.39% to 0.51%, and reduced unsubscribes from 1.2% to 1.0%; related experiments across three firms also reported benefits.Evidence A · P1
  292. MWITA-ELO-2026-002Two randomized German field experiments did not reproduce a benefit from first-name subject-line personalization among students: detected opens were 3.2% lower and clicks similar, both nonsignificant; title-plus-surname personalization among politicians raised opens 11.8% significantly while its 14% click lift was nonsignificant.Evidence A · P1
  293. MWITA-ELO-2026-003Across three longitudinal field experiments, emails tailored with data applicants had actively volunteered outperformed observed-data personalization in aggregate open, click-to-open and conversion outcomes, but unsubscribe differences were not significant and ordinary observed-data personalization did not outperform the untailored control.Evidence A · P1
  294. MWITA-ELO-2026-004Reducing coupon-email frequency from daily to every other day for one week cut unsubscribe probability 59.2% but also reduced daily revenue 7.6%, purchase probability 4.8% and item quantity 10%; price per item was effectively unchanged.Evidence A · P1
  295. MWITA-ELO-2026-005Two incremental engagement emails over one month increased five-month subscriber retention likelihood 7.4%, reduced churn odds 26.3% and increased per-period service consumption 7.0%, yet modeled profit rose only for some subscription/use segments and fell for others because retained usage creates service cost.Evidence A · P1
  296. MWITA-ELO-2026-006In the stratified randomized beauty-retail experiment, email had no significant sales effect in any of six customer-value segments, despite observational models finding email responsiveness for some current-customer segments; direct mail worked for prospects and the email-plus-mail interaction was small but significant only for medium-value customers.Evidence A · P1
  297. MWITA-ELO-2026-007Across 70 randomized experiments on a ticket-resale platform, emailed discounts increased average expenditure by $3.03, or 37.2%, during the promotion window and by $1.55 the following week; about 90% of the promotion-window gain occurred without offer redemption.Evidence A · P1
  298. MWITA-ELO-2026-008Varying the subject line across weekly lifecycle emails produced small detected-open gains: 1.1 percentage points across weeks 2-3 among active/new learners and 0.3 points among all new learners; the fixed-effects meta-analysis estimated +0.7 points, while course progress was null.Evidence A · P1
  299. MWITA-ELO-2026-009Across three randomized field experiments, authority cues in sender names or subject lines sometimes improved early detected opens but did not improve clicks; the initially selected subject-line winner was not statistically significant when measured at 30 days, indicating a winner-timing/fade-out problem.Evidence A · P1
  300. MWITA-ELO-2026-010Adding problem-solving content to four monthly emails reduced rather than increased access to an audit-and-feedback tool by 12.9%, while anticipated-regret and material-incentive content did not significantly increase access; fewer than half opened and fewer than one in ten clicked.Evidence A · P1
  301. MWITA-ELO-2026-011Among people who had enrolled but not activated an e-cohort, three reminders mixing standard email, SMS and institutional email increased 30-day questionnaire completion 27% relative to a single-mode approach; SMS accelerated response and institutional/formal email appeared more effective than standard wording.Evidence A · P1
  302. MWITA-ELO-2026-012Measurement of the Alexa Top Million and more than a year of Gmail SMTP connections found that among a long tail of over 700,000 SMTP servers, only 35% successfully configured encryption and 1.1% published a DMARC authentication policy; in seven countries, over 20% of inbound Gmail messages arrived in cleartext due to observed network attacks.Evidence A · P1
  303. MWITA-LRO-2026-003Hotels that began publicly responding to reviews gained about 0.12 stars and 12% more review volume, while their remaining one- and two-star reviews became about 10% longer.Evidence B · P1
  304. MWITA-LRO-2026-004Roughly 16% of sampled Yelp restaurant reviews were filtered as suspicious; filtering was more likely when reputation was weak, after bad reviews, and under stronger competition.Evidence B · P1
  305. MWITA-LRO-2026-009Randomly soliciting a hotel review reduced the solicited customer's subsequent brand spending by $16, or 7%, on average; harm was concentrated among less-than-perfectly-satisfied, highly loyal customers.Evidence A · P1
  306. MWITA-LRO-2026-012Without prior reviews, Airbnb requests from guests with African American-sounding names were accepted 19.2 percentage points less often; one credible peer review made Black–White acceptance rates statistically indistinguishable.Evidence A · P1
  307. MWITA-CUX-2026-001In randomized holdbacks on Stripe Checkout, surfacing at least one additional locally relevant payment method beyond cards increased completed purchase conversion by 7.4% and revenue by 12% on average; effects were highly heterogeneous by method and market.Evidence A · P1
  308. MWITA-CUX-2026-002In a 50/50 randomized holdback over more than 150,000 eligible Checkout sessions, displaying one eligible BNPL method increased conversion across order values and produced up to 14% more revenue; Stripe attributes more than two thirds of BNPL volume to incremental rather than displaced purchases.Evidence A · P1
  309. MWITA-CUX-2026-003Across 1.5 million subscription Checkout sessions with a randomized 1% holdback, presenting localized currency through Adaptive Pricing increased signup conversion by 4.7%, authorization by 1.9% and subscription lifetime value per session by 5.4%.Evidence A · P1
  310. MWITA-CUX-2026-004A large StubHub field experiment found that users shown mandatory fees only late in checkout were 13% more likely to purchase and spent about 21% more than users shown all-in prices upfront, while spending 5.42% more conditional on purchase; the revenue gain therefore coexisted with reduced price transparency.Evidence A · P1
  311. MWITA-CUX-2026-005In a randomized German cinema checkout experiment with more than 34,000 customers, shrouding a EUR 3 3D surcharge increased initiation of 3D purchases but also increased checkout dropout, leaving final 3D demand statistically unchanged across disclosure conditions.Evidence A · P1
  312. MWITA-CUX-2026-006Among 2,605 recruitment-site visitors, one infographic landing page reduced interest-form completion from 16.8% to 12.8% (OR 0.73, p=.019), while a second infographic was null; neither changed eventual study randomization.Evidence A · P1
  313. MWITA-CUX-2026-007Two randomized web experiments found that shorter commercial landing pages generated significantly more lead-form submissions than longer information-rich versions across locations and observation periods.Evidence A · P1
  314. MWITA-CUX-2026-008In a randomized eight-condition landing-page field experiment with 956 visitors, repeating the headline phrase verbatim on the call-to-action button increased click-through by more than 10 percentage points relative to introducing new wording.Evidence A · P1
  315. MWITA-CUX-2026-009In a randomized federal digital-form test, embedding instructions at the relevant fields raised submission from 33.0% to 36.2%, a 3.2 percentage-point estimate whose 95% confidence interval included zero.Evidence A · P1
  316. MWITA-CUX-2026-010In a preregistered-style online choice experiment with 5,889 UK participants, preselecting shipping or protection add-ons made consumers about 60%-70% more likely to select the more expensive option, whereas merely visually mimicking preselection without an actual default was null.Evidence A · P1
  317. MWITA-CUX-2026-011In a six-country randomized online experiment with 7,430 participants, tested dark-pattern interfaces increased inconsistent choices by 6-12 percentage points versus placebo interfaces, and a transparency remedy did not eliminate the distortion.Evidence A · P1
  318. MWITA-CUX-2026-012Among 75,914 new ecommerce visitors, requesting registration before browsing increased registrations by 58.08% but left 24-hour purchase statistically unchanged; among users induced to register, instrumental-variable estimates over 47 days indicated higher purchase probability, order count and revenue.Evidence A · P1
  319. MWITA-NCO-2026-006Immersive VR increased self-reported enjoyment and attention for an art program but did not improve immediate factual retention over iPad delivery.Evidence A · P1
  320. MWITA-NCO-2026-009An unlabeled political deepfake reduced candidate support, credulous comments worsened the loss, and an incredulous comment made the deepfake effect statistically indistinguishable from baseline.Evidence A · P1