Evidence infrastructure
Sources & methodology
Every regulated claim on Disclara must resolve to a structured source record with its issuer, authority type, canonical URL, relevant dates, status, reviewed snapshot marker and review owner.
What the registry makes visible
Primary sources
Binding law, regulator guidance, first-party platform policy and technical standards are preferred.
Effective windows
Publication, effective-from and effective-to dates are separated; checker decisions use the planned publication date.
Freshness
Each source carries a last-checked date, status and reviewed snapshot marker.
Human accountability
Important guidance pages display the review owner, role and review date.
Conflict detection
Simultaneous web, PDF or API versions are compared; material conflicts suspend automatic interpretation.
Source model
source_id— stable identifier referenced by rules, pages and changelog entries.issuerandsource_type— law, guidance, platform policy, standard, first-party announcement or secondary report.canonical_url— the underlying source rather than an uncited summary.publication_date,effective_fromandeffective_to— when a source appeared and its known applicability window.last_checked_atandstatus— when the source was inspected and whether it is current, scheduled, superseded or uncertain.content_hashandreviewed_by— the reviewed snapshot marker and accountable editorial byline.source_version_id, format and retrieval time — the exact observed surface behind a snapshot digest.conflict_statusand authority rank — whether materially different versions require human review.
Current entries marked manual:YYYY-MM-DD use a manual review marker, not a cryptographic copy of the external page. The field can hold a content digest when automated source snapshots are introduced.
Fail-closed interpretation
Source Conflict Sentinel
Explore the structured Rule Graph: rules, dependencies and platform labels →
When simultaneous versions from the same authority can change the decision, Disclara records the conflict and stops automatic approval. The public conflict register shows the exact version IDs, formats, observation dates and SHA-256 digests.
Editorial rules
- Regulatory pages cite primary legal or regulatory sources.
- Platform pages cite first-party platform policies wherever available.
- No detected diff becomes legal interpretation automatically; material changes require review.
- Uncertainty remains visible and may produce “Human review required”.
- Disclara never presents a result as guaranteed compliance or legal approval.
- Reviewer qualifications appear only after the identity, consent and credential are verified.
81 structured records
Current source registry
ph_asc_circular_2026_014acurrentThe PDF is dated October 1; the web page was published October 2. No separate new effective date is stated. Disclara's October 2 cutover is editorial, with October 1 requiring date review. Disclosure depends on an AI-changed element's underlying documentation requirement; the real-child DOLE example is counterfactual, not a blanket synthetic-child permit requirement. Visual font/size constraints and child-specific frames/layouts apply. Registry metadata is not a retained source body.
anthropic_claude_marking_2026currentDetected text marks indicate possible generation or processing, not original authorship. Non-detection does not exclude AI. Detector access is private preview for eligible organisations; support varies by model and surface. Disclara records reports but has no detector integration.
in_asci_sgc_guidelines_2026currentFinal ASCI self-regulatory guideline, published 29 September 2026 and effective on expiration of three months. The 17 September chairman signature is not the publication date. Consumer-material synthetic depictions and sponsored AI product suggestions are separate branches; a suitable platform label may implement the AI-origin disclosure. Consent contract granularity is an operational control, not a verbatim ASCI clause.
adobe_premiere_content_credentials_2026currentAdobe documents credential inclusion on supported exports containing Adobe AI-generated assets. This does not establish that any audio cleanup is synthetic speech, nor guarantee a downstream platform label.
tiktok_shop_us_ai_listing_images_2026currentUS product-listing image policy: AI depiction must remain consistent with the selected SKU, PDP and delivered product. Publication date is not a separately confirmed effective date; no retroactive rule or applicability to other markets is inferred.
tiktok_shop_us_ai_content_2026currentUS Shop policy distinguishes AI-use disclosure for fully/significantly generated or altered content from accurate product presentation. An AI label does not resolve product mismatches.
iab_austria_ai_labelling_2026currentPublic summary only. It says allocation among agency, advertiser and publisher depends on who decides and influences AI use, recommends contractual clarity, and requires case-specific assessment. This is industry guidance, not binding law; the downloadable five-page guide was not independently verified for precise examples.
apple_reference_image_security_2026currentReference Image attests sensor capture and a reference image, not the truth of the depicted scene or photographer identity. Disclara does not currently parse or verify Apple Reference Image data.
apple_reference_image_support_2026currentSharing/export can omit the reference data; Apple currently documents viewing on iOS, iPadOS and macOS 27 or later.
apple_iphone_18_pro_reference_image_2026currentApple says Reference Image capture will not be available in the EU at launch, although iOS/iPadOS/macOS 27 users there can develop and view reference images.
stanford_marketing_ai_guidelines_2026currentUniversity-specific guidance requires rights clearances and prohibits false AI depictions of Stanford people, events, research, facilities and achievements. It is an example of a customer-specific policy boundary, not a default Disclara rule for other organisations.
ph_asc_circular_2026_014currentSections III-IV distinguish broad AI use in production from disclosure for advertisements containing content created, generated or materially enhanced by AI. No numerical materiality threshold is supplied. The circular is ASC self-regulation, not a general statute.
belgian_advertising_code_2026uncertainThe official web and PDF versions contain materially different Article 9.1 AI-disclosure thresholds as of 19 September 2026. The active conflict is recorded separately and must not be silently resolved by the checker.
uba_belgian_advertising_code_launch_2026currentUBA recommends the online Code as the starting point because versions may be updated. That recommendation does not by itself resolve a material conflict between simultaneous official web and PDF wording.
meta_oversight_board_ai_video_uk_politician_2026currentThe decision records that no AI Info label was applied where the uploader did not self-disclose and recognized indicators were absent. It also reports limited at-scale enforcement. Label absence therefore cannot establish that content is not AI-generated.
eu_ucpd_2005_29_consolidatedcurrentAnnex I points 23b and 23c address claims that reviews come from actual consumers, false consumer reviews or endorsements, and misrepresentation of reviews or endorsements. This source does not decide an asset-specific deception question without its full context.
eu_ucpd_consumer_reviews_guidancecurrentSection 4.2.4 explains consumer-review and endorsement practices, including that presentation is assessed from the average consumer's perspective rather than wording alone. Guidance is interpretive and is not an asset-specific legal determination.
ibc_c2pa_stamping_recovery_2025currentIBC describes an accelerator proof of concept for organization-authorized publication stamping, verification and recovery of stripped manifests through soft bindings. It does not establish general product availability or guaranteed recovery.
pinterest_gen_ai_labelscurrentPinterest documents AI-modified labels for Pins and a separate ad disclosure surface based on owner input or platform detection.
pinterest_policy_update_2026_11_12scheduledFirst-party notice that updated Terms, Advertising Guidelines, enforcement materials and GenAI Acceptable Use Guidelines take effect on 12 November 2026; each final policy must be rechecked at effectiveness.
guardian_influencer_deepfake_ads_2026currentSecondary reporting illustrating the operational need to preserve authorization-scope and incident evidence; it is not a legal authority or proof of platform response times in every case.
microsoft_advertising_agreement_2025currentThe agreement keeps responsibility for ad content and applicable law with the advertiser, permits Microsoft to reformat or modify ads, and states that non-enforcement does not waive Microsoft's rights. It does not create a universal AI-disclosure duty for every ad.
microsoft_ads_ai_generated_advertising_guide_2026uncertainRechecked 23 September 2026: the first-party help URL still returns an 'Oops' / no matching result page. It cannot support detailed format-specific AI-disclosure claims until Microsoft restores or replaces it.
microsoft_ads_copilot_provenancecurrentThe current English page describes AI-generated advertising assets and variations. It does not establish that every Microsoft Advertising output carries C2PA or a watermark, or that every ad format has a viewer-visible AI disclaimer.
timberborn_platform_generated_ad_variants_2026currentA first-party developer comment in a community thread reports that an approved trailer was transformed into AI-generated ad variants without those variants being shown for approval. It is operational evidence for approved-versus-served controls, not a general finding about all platform delivery systems.
observed_ai_ad_enforcement_thread_2026uncertainAnecdotal reports of limited rejection are observations about enforcement, not authoritative policy evidence and not a basis for lowering a legal or platform requirement.
midjourney_video_docscurrentmidjourney_editor_docscurrentadobe_firefly_content_credentialscurrentopenai_chatgpt_image_provenancecurrentCompatible images generated with ChatGPT, Codex and the OpenAI API include C2PA metadata and SynthID. Metadata may be removed by later processing or screenshots.
openai_sora_safety_2026currentelevenlabs_audio_watermarkingcurrentelevenlabs_safetycurrentElevenLabs describes embedded watermarking and support for provenance standards including C2PA; this does not establish that every export contains a C2PA credential.
heygen_create_video_apicurrentrunway_image_to_videocurrentcanva_magic_mediacurrentcanva_ai_product_terms_2026currentCanva prohibits removing, altering or disabling provenance or metadata tags from AI-generated content; this does not establish C2PA on every export.
google_gemini_ai_verificationcurrentlinkedin_content_credentialscurrentGradual C2PA label support for signed image and video; sponsored single-image and video ads require direct media upload for the icon to appear.
linkedin_ai_content_best_practicescurrentgoogle_ads_ai_labelcurrentAdvertiser AI-label control, My Ad Centre disclosure and visible overlays for designated geographies including campaigns targeting the EU.
x_authenticity_policycurrentx_content_noticescurrentX may apply a Manipulated media notice; this is not a general creator-controlled AI disclosure toggle.
ec_article_50_guidelines_2026currentec_article_50_faq_2026currentThe Commission distinguishes provider-side machine-readable marking from deployer-side human-visible disclosure, excludes personal non-professional use, explains that employees under a legal person's control are not separate deployers, and records the Article 50(2) grace period until 2 December 2026 for systems placed on the market before 2 August 2026.
gamesbeat_twitch_ai_disclosure_interview_2026currentSecondary reporting of a direct CEO interview. It supports a future Twitch capability entry but is not itself a Twitch policy document and must not be treated as proof that no external disclosure duty applies.
ec_ai_content_code_2026currentVoluntary code assessed by the Commission and AI Board as an adequate means to demonstrate compliance; adherence is not conclusive evidence of compliance.
ec_eu_ai_content_icons_2026currentThree optional icons in black, white and 50% transparency variants; freely usable without attribution. Icon use alone does not establish legal compliance.
eurlex_ai_act_2024_1689currentArticle 50 transparency obligations apply from 2 August 2026 (Art. 113).
ec_ai_act_frameworkcurrenttiktok_aigc_supportcurrenttiktok_integrity_authenticitycurrentmeta_labeling_ai_contentcurrentmeta_newsroom_labeling_2024currentmeta_ads_ai_transparency_2026currentUpdated 1 June 2026. About this ad rollout and label presentation vary by region; not an account-level UI test.
youtube_altered_content_helpcurrentyoutube_blog_disclosure_2024currentc2pa_speccurrentcipa_exif_3_1_2026currentCIPA lists the current Exif standard. Disclara's browser preflight inspects a bounded subset of standard TIFF/Exif tag identifiers, not their values, MakerNotes, XMP privacy fields or all possible metadata.
adobe_apsb26_147_content_credentials_sdkcurrentAdobe lists c2pa-v0.89.2 and earlier and c2patool-v0.26.70 and earlier as affected; fixed releases are c2pa-v0.90.17 and c2patool-v0.27.17. The bulletin covers security-feature bypass and denial of service, and says Adobe knows of no exploitation in the wild as of publication. A version check against this bulletin is not a full security assessment.
c2pa_js_web_0_15_1_releasecurrentPinned browser SDK release with c2pa-wasm 0.13.0. Its release commit 38cf9cd pins c2pa-rs 0.90.22 in Cargo.lock. Disclara disables remote manifest fetching, OCSP and trust verification in its local integrity-only preflight; these boundaries are application configuration, not a claim of general verifier security.
iptc_synthetic_media_guidancecurrentDescribes IPTC Digital Source Type values for fully algorithmic and composite synthetic media.
murphycobb_amplify_2026currentDescribes technology-independent validation against cost, speed, capacity, quality, risk and adoption, followed by workflow, technology, governance and partner implementation. It is market evidence, not a regulatory source.
fotoware_content_authenticity_2026currentDescribes preservation and inspection of C2PA provenance from ingestion through edits and third-party systems to publication.
bynder_content_provenance_2026currentDocuments C2PA preservation on original assets and manually uploaded derivatives, plus C2PA and SynthID on GenAI Smart Edit variations.
c2pa_orgcurrentcontent_credentials_verifycurrentThe public verifier currently redirects to verify.contentauthenticity.org and requires JavaScript for file inspection.
c2pa_implementation_guide_2026currentIntroduces implementation material for digitalSourceType, c2pa.ai-disclosure, Regions of Interest, inputTo ingredients and action-level AI usage.
c2pa_synthetic_content_whitepaper_2026currentc2pa_deployment_guidance_2026currentPractical guidance for selecting, generating, preserving, verifying and displaying Content Credentials, including valid, trusted and not-found results.
wordpress_image_strip_metacurrentWordPress strips metadata from resized images by default; the documented filter applies to Imagick, while GD always strips profiles.
meta_c2pa_ai_generated_imagescurrentMeta documents detecting C2PA and IPTC AI-generation signals while acknowledging that invisible markers can be stripped.
adobe_photoshop_content_credentials_exportcurrentAdobe documents cloud and embedded export modes and warns that attached credentials can be stripped after publication.
adobe_premiere_content_credentialscurrentAdobe documents format, export, preferences, identity and destination-support dependencies separately from asset content.
whatsapp_channel_ai_labelcurrentChannel-specific AI content label; availability and automatic application may vary.
whatsapp_channel_ai_label_rolloutcurrentSecondary rollout evidence for label permanence, Channel scope and phased availability; verify against first-party product behavior.
iab_ai_transparency_framework_2026currentIndustry guidance for synthetic voice, digital twins, consumer-facing disclosure and machine-readable provenance; not binding law.
c2pa_ux_guidance_2_2currentDistinguishes full generation from partial AI editing and warns that a generic Made with AI label can misrepresent light or localised edits.
california_sb_1050_2026scheduledChapter 246, Statutes of 2026. Applies to advertisements that prominently include a synthetic performer; the enacted text contains defined exclusions.
california_constitution_article_iv_8currentSection 8(c) supplies the general effective-date rule for non-urgency statutes enacted in a regular session.
iab_nz_ai_transparency_framework_v2_2026currentIndustry guidance supporting a materiality-driven approach and avoiding unnecessary labels and disclosure fatigue; not binding law.
See the rule changelog for dated policy updates and affected pages. Information on this site is not legal advice.