Wetin you go tell client about EU AI Act Article 50 label
Commission Article 50 page separate provider duty from deployer duty for interaction notice, machine-readable mark, and deep-fake disclosure (https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50). One law-firm alert talk say those duty become applicable on 2 August 2026 and dem no defer am with high-risk rule (https://www.goodwinlaw.com/en/insights/publications/2026/08/alerts-technology-dpc-eu-ai-act-transparency-obligations-now-in-force). General information, e no be legal advice.
Dis article na machine translation.
2026-10-08. Dis na general information for consultants, e no be legal advice. Make counsel read the regulation before client rely on am. Dem use two public page. The European Commission Article 50 page na https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50 . The law-firm alert na https://www.goodwinlaw.com/en/insights/publications/2026/08/alerts-technology-dpc-eu-ai-act-transparency-obligations-now-in-force . The alert talk say e no constitute legal advice. Date, the regulation number, and the fine figure wey dey only for the alert na that firm own account. Dem no retrieve the official EUR-Lex text for dis check.
The Commission page talk say the text wey dey show there base on the EUR-Lex consolidated version of the AI Act as at 27 July 2026. E talk say, where dem show am, “NEW” and “AMENDED” identify addition or amendment wey Digital Omnibus on AI do. E talk say the summaries “are not legal binding.” The page no state the date wey Article 50 become applicable. That date dey only for the law-firm section below.
## Provider and deployer no be the same duty
The page give some duty to provider and some to deployer. The passage wey dem use here no define those word. Client job title no decide the role. Counsel suppose read the regulation own definition before anybody tell person which paragraph apply.
## Wetin the Commission page talk say provider must do
Paragraph 1. Provider shall make sure say AI system wey dem intend to interact direct with natural person dey designed and developed so that the person wey e concern go know say dem dey interact with AI system, unless e clear from the view of natural person wey get reasonable information, wey dey observant and circumspect, as dem consider the circumstance and the context of use. The page talk say dis obligation no go apply to AI system wey law authorise to detect, prevent, investigate or prosecute criminal offence, subject to appropriate safeguard for the right and freedom of third party, unless those system dey available for the public to report criminal offence.
Paragraph 2. Provider of AI system, including general-purpose AI system, wey dey generate synthetic audio, image, video or text content, shall make sure say the output dey marked for machine-readable format and person fit detect am as artificially generated or manipulated. The page talk say the technical solution suppose dey effective, interoperable, robust and reliable as far as e technically feasible, as dem consider the specificity and limitation of the type of content, the cost of implementation and the generally acknowledged state of the art, as relevant technical standard fit show. E talk say dis obligation no go apply to the extent wey the system dey do assistive function for standard editing, or dem no substantially alter the input data wey the deployer give or the semantics, or where law authorise am to detect, prevent, investigate or prosecute criminal offence.
## Wetin the Commission page talk say deployer must do
Paragraph 3. Deployer of emotion recognition system or biometric categorisation system shall inform the natural person wey e expose to the operation of the system, and shall process the personal data in accordance with Regulations (EU) 2016/679 and (EU) 2018/1725 and Directive (EU) 2016/680, as e apply. The page talk say dis obligation no go apply to AI system wey dem use for biometric categorisation and emotion recognition wey law permit to detect, prevent or investigate criminal offence, subject to appropriate safeguard for the right and freedom of third party, and in accordance with Union law.
Paragraph 4, image, audio or video. Deployer of AI system wey generate or manipulate image, audio or video content wey constitute deep fake shall disclose say the content don artificially generate or manipulate. The page use the word “deep fake.” Dis article no restate definition, because the definition no dey for the passage wey dem use here. The page talk say dis obligation no go apply where the use dey authorised by law to detect, prevent, investigate or prosecute criminal offence. Where the content dey part of evidently artistic, creative, satirical, fictional or analogous work or programme, the page limit that transparency obligation to disclosure of the existence of such generated or manipulated content for appropriate manner wey no hamper the display or enjoyment of the work.
Paragraph 4, text. Deployer of AI system wey generate or manipulate text wey dem publish with the purpose to inform the public on matter of public interest shall disclose say the text don artificially generate or manipulate. The page talk say dis obligation no go apply where the use dey authorised by law to detect, prevent, investigate or prosecute criminal offence, or where the AI-generated content don undergo human review or editorial control and natural or legal person hold editorial responsibility for the publication of the content.
Paragraphs 5 to 7. The information for paragraph 1 to 4 shall reach the natural person wey e concern for clear and distinguishable manner at the latest at the time of the first interaction or exposure, and e shall conform to the applicable accessibility requirement. Paragraph 1 to 4 no go affect the requirement and obligation for Chapter III, and dem no go prejudice other transparency obligation wey Union or national law lay down for deployer of AI system. Paragraph 7 dey marked AMENDED. The Commission shall encourage and facilitate code of practice for Union level for the detection, marking and labelling of artificially generated or manipulated content. As dem take utmost account of the opinion of the Board, the Commission shall assess whether adherence to those code dey adequate to ensure compliance with paragraph 2 and 4, in accordance with the procedure for Article 56(6). If e deem code inadequate, e fit adopt implementing act under the examination procedure for Article 98(2). The page list recitals 132, 133, 134, 136 and 137. Dis article no quote those recital.
## The date, and why high-risk date move, na the law firm own account
Wetin follow na the law firm own account, e no be reading of EUR-Lex. The alert talk say on 2 August 2026 the transparency and information obligation under Article 50 become generally applicable and national competent authority for the whole EU fit enforce am. E talk say the duty apply whether or not system dey classified as high-risk, with one narrow exception wey relate to the labelling of legacy generative system. The Commission page wey dem use above no state that date.
The alert talk say 2 August 2026 na also the date wey dem originally bill for the high-risk provision, and say that one no happen. E talk say the AI Omnibus, wey e cite as Regulation (EU) 2026/1744 and say e come into force on 27 July 2026, push those obligation back. Dem no retrieve the official text of that regulation here, so the regulation number and that in-force date na the firm own citation. The same calendar day, 27 July 2026, na also the date wey the Commission page give for the consolidated version wey e display. Those na two different statement. The alert talk say Annex III standalone high-risk system, including those wey dem use for employment, education, law enforcement and critical infrastructure, go apply from 2 December 2027. E talk say AI system wey dey inside product wey EU product-safety legislation regulate (Annex I), such as medical device, toy and lift, go apply from 2 August 2028. E talk say machinery dey treated different: the AI Act requirement go instead dey transposed into the machinery regulation by delegated act wey due by 2 August 2028.
The alert talk say those extension no affect the prohibition on unacceptable AI practice, the obligation wey relate to general-purpose AI model, or the transparency obligation, all of which e talk say continue to apply under the existing timetable. E talk say provider of generative system wey already dey market before 2 August 2026 get until 2 December 2026 to comply with the marking obligation under the AI Omnibus. E talk say the Commission adopt guideline on the transparency obligation on 20 July 2026. Dem no retrieve those guideline for dis check.
The alert talk say non-compliance fit give rise to administrative fine of up to €15 million or 3% of worldwide annual turnover, whichever higher, and e cite Article 99(4). Dem no retrieve Article 99. No quote that figure to client as figure wey dem don check against the regulation. The Commission page content list get entry wey title na “Article 99: Penalties,” and that list no be the penalty text.
The alert shorter summary talk say provider and deployer of system wey dey generate synthetic audio, image, video or text, including wetin e call deepfake, must mark output for machine-readable, detectable format, and say deployer must also disclose deepfake wey depict real person, place or event. The Commission page put the machine-readable marking duty on provider, for paragraph 2, and the deep-fake disclosure duty on deployer, for paragraph 4. Paragraph 4 for that page no use the phrase “real persons, places, or events.” Use the Commission page for who must mark and who must disclose. Treat the alert shorter summary as the firm own compression, no be the article text.
## File wey client talk say na deepfake
Keep the file as e land. Note the time, the channel, and whether any disclosure or mark dey visible for the first moment wey the person see am. No edit the file to add label, remove label, or change label. Dis article no describe how label dey made, how detector dey work, or how person fit avoid label.
Label wey no dey no be proof say the content genuine. Label wey dey no be proof of who send the file, and e no be way to recover money. Dis article no set out reporting procedure. Counsel na the one wey go name the authority, if any dey. Nothing here promise say record go get loss back.
## Checklist wey no be legal advice
- Read the Commission Article 50 page with counsel, including the exception, before you tell client which duty apply.
- Ask whether the client provide system wey people interact with, provide synthetic audio, image, video or text, deploy deep fake, deploy public-interest text, or deploy emotion recognition or biometric categorisation. The page give those situation different duty.
- No assume exception. The page exception include interaction wey dey obvious to the person wey paragraph 1 describe, assistive function for standard editing, no substantial change to the deployer input or e semantics, evidently artistic, creative, satirical, fictional or analogous work, human review or editorial control with editorial responsibility, and authorisation by law for criminal offence.
- Until counsel check the official text, treat the 2 August 2026 application date, the high-risk deferral, the legacy marking window to 2 December 2026, the guidelines date of 20 July 2026, Regulation (EU) 2026/1744, and the fine figure as the law firm own account.
- If client bring file wey dem believe na synthetic, preserve am, record when and where e land, and record whether disclosure dey visible. No alter am.
- No promise say label, missing label, or saved note go recover loss.
Further note for dis audience dey for the consultants hub: https://blog.mithril.fund/for/consultants .
## Wetin Mithril go do and wetin e no go do
The public docs describe Your records (あなたの記録) as the place wey open after sign-in, where person choose category, write wetin happen, and add note: https://docs.mithril.fund/app-guide . That guide talk say dem no record the screen after login, so dis article no describe the layout. Dis article call that signed-in recording path Live only for the sense say the public guidance present am as wetin open after sign-in. Hosted DAST and SAST na Planned. The security-services page talk say dem dey gated, dem never ready as hosted service, and dem na planned work: https://docs.mithril.fund/reference/security-services . Mithril no investigate file, no name who send am, no recover money, and no give legal advice. No other product status dey stated here.