EU AI Act

Properties

jurisdiction: European Union
level: supranational
domain: general
status: in-effect
scope: Horizontal, risk-based rules for AI systems placed on or used in the EU market, tiered from prohibited to high-risk to limited and minimal risk, with a separate regime for general-purpose AI models.
obligations: Risk classification; high-risk systems need risk management, data governance, technical documentation, logging, transparency, human oversight and conformity assessment; GPAI models need technical documentation and (for systemic-risk models) evaluation and incident reporting.
effectiveDate: 2024-08-01
enactedDate: 2024-06-13
penalties: Up to 35,000,000 EUR or 7% of global annual turnover for prohibited-practice breaches.
appliesTo: providers, deployers, importers and distributors of AI systems in the EU
officialUrl: https://eur-lex.europa.eu/eli/reg/2024/1689/oj
notes: Phased application (original): prohibitions from Feb 2025, GPAI obligations from Aug 2025, most high-risk obligations from Aug 2026, embedded high-risk from Aug 2027. UPDATE (Digital Omnibus on AI): Regulation (EU) 2026/1744, amending the AI Act (Regulation (EU) 2024/1689), was signed 8 July 2026 and published in the Official Journal (L series) on 24 July 2026; per its own entry-into-force clause it took effect on the third day after publication, 27 July 2026 (independently reconfirmed 2026-09-07 via EUR-Lex search results and the European Commission's own AI Act policy page, which also states entry into force was 27 July 2026). Confirmed against the consolidated OJ text: stand-alone high-risk (Annex III) AI Act obligations are deferred to 2 December 2027, and embedded high-risk (Annex I, product-safety-linked) obligations to 2 August 2028. The general Article 50 AI-generated-content transparency obligation applied as scheduled from 2 August 2026; this specific date is confirmed this round directly on the European Commission's own AI Act policy page (digital-strategy.ec.europa.eu). Article 5(1) new prohibitions on AI-generated non-consensual intimate imagery ('nudification' apps) and CSAM: the substantive text is independently confirmed verbatim against the official EUR-Lex consolidated text (CELEX 02024R1689, consolidated version dated 27.07.2026): point (ba) reads 'the placing on the market, the putting into service or the use of an AI system that generates or manipulates realistic images, videos, audio or similar material of an identifiable natural person's intimate parts, or of an identifiable natural person engaged in sexually explicit activities, without that person's freely-given, specific, informed, unambiguous and explicit consent for that generation or manipulation'; point (bb) separately covers AI-generated material within the meaning of Directive 2011/93/EU (CSAM). The AI Act itself remains in force; effectiveDate and status fields are unchanged. UPDATE (this run, 2026-09-19, confirmed directly against the official Official Journal PDF text of Regulation (EU) 2026/1744, OJ L, 24.7.2026, ELI: http://data.europa.eu/eli/reg/2026/1744/oj, downloaded from eur-lex.europa.eu, and independently cross-checked via search): both open sub-date questions flagged in prior rounds are now resolved from the amending regulation's own operative text, not a secondary paraphrase. First, amendment (39)(b) inserts a new Article 111(4) into the AI Act reading verbatim: 'Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, that have been placed on the market before 2 August 2026 shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026.' This confirms the Article 50(2) watermarking/machine-readable-marking grace period to 2 December 2026 that a prior round could not re-locate on the Commission's policy page; it is enacted via new Article 111(4), not via Article 113, which explains why it was not found under Article 113/Article 50 lookups. This retracts the prior round's flag that the grace period should be treated as unconfirmed: it is now confirmed to the day via the primary Official Journal text. Second, amendment (40)(a) replaces Article 113, third paragraph, point (a) to read verbatim: 'Chapters I and II shall apply from 2 February 2025, with the exception of Article 5(1), first subparagraph, points (ba) and (bb), and Article 5(1a) and (1b) which shall apply from 2 December 2026;' This confirms the exact day (2 December 2026, not merely 'December 2026') for the Article 5(1) prohibitions on AI-generated non-consensual intimate imagery ('nudification') and CSAM (Prohibition 9), retracting the prior round's flag that only the month, not the day, was confirmed. Both quotes were read directly from the OJ PDF's own consecutively numbered amendment clauses (39) and (40) to Article 1 of Regulation (EU) 2026/1744, on page 35 of 41 of the official PDF, and independently corroborated via search results quoting the same amendment clauses verbatim.

Properties

jurisdiction

European Union

level

supranational

domain

general

status

in-effect

scope

Horizontal, risk-based rules for AI systems placed on or used in the EU market, tiered from prohibited to high-risk to limited and minimal risk, with a separate regime for general-purpose AI models.

obligations

Risk classification; high-risk systems need risk management, data governance, technical documentation, logging, transparency, human oversight and conformity assessment; GPAI models need technical documentation and (for systemic-risk models) evaluation and incident reporting.

effectiveDate

2024-08-01

enactedDate

2024-06-13

penalties

Up to 35,000,000 EUR or 7% of global annual turnover for prohibited-practice breaches.

appliesTo

providers, deployers, importers and distributors of AI systems in the EU

officialUrl

https://eur-lex.europa.eu/eli/reg/2024/1689/oj

notes

Phased application (original): prohibitions from Feb 2025, GPAI obligations from Aug 2025, most high-risk obligations from Aug 2026, embedded high-risk from Aug 2027. UPDATE (Digital Omnibus on AI): Regulation (EU) 2026/1744, amending the AI Act (Regulation (EU) 2024/1689), was signed 8 July 2026 and published in the Official Journal (L series) on 24 July 2026; per its own entry-into-force clause it took effect on the third day after publication, 27 July 2026 (independently reconfirmed 2026-09-07 via EUR-Lex search results and the European Commission's own AI Act policy page, which also states entry into force was 27 July 2026). Confirmed against the consolidated OJ text: stand-alone high-risk (Annex III) AI Act obligations are deferred to 2 December 2027, and embedded high-risk (Annex I, product-safety-linked) obligations to 2 August 2028. The general Article 50 AI-generated-content transparency obligation applied as scheduled from 2 August 2026; this specific date is confirmed this round directly on the European Commission's own AI Act policy page (digital-strategy.ec.europa.eu). Article 5(1) new prohibitions on AI-generated non-consensual intimate imagery ('nudification' apps) and CSAM: the substantive text is independently confirmed verbatim against the official EUR-Lex consolidated text (CELEX 02024R1689, consolidated version dated 27.07.2026): point (ba) reads 'the placing on the market, the putting into service or the use of an AI system that generates or manipulates realistic images, videos, audio or similar material of an identifiable natural person's intimate parts, or of an identifiable natural person engaged in sexually explicit activities, without that person's freely-given, specific, informed, unambiguous and explicit consent for that generation or manipulation'; point (bb) separately covers AI-generated material within the meaning of Directive 2011/93/EU (CSAM). The AI Act itself remains in force; effectiveDate and status fields are unchanged. UPDATE (this run, 2026-09-19, confirmed directly against the official Official Journal PDF text of Regulation (EU) 2026/1744, OJ L, 24.7.2026, ELI: http://data.europa.eu/eli/reg/2026/1744/oj, downloaded from eur-lex.europa.eu, and independently cross-checked via search): both open sub-date questions flagged in prior rounds are now resolved from the amending regulation's own operative text, not a secondary paraphrase. First, amendment (39)(b) inserts a new Article 111(4) into the AI Act reading verbatim: 'Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, that have been placed on the market before 2 August 2026 shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026.' This confirms the Article 50(2) watermarking/machine-readable-marking grace period to 2 December 2026 that a prior round could not re-locate on the Commission's policy page; it is enacted via new Article 111(4), not via Article 113, which explains why it was not found under Article 113/Article 50 lookups. This retracts the prior round's flag that the grace period should be treated as unconfirmed: it is now confirmed to the day via the primary Official Journal text. Second, amendment (40)(a) replaces Article 113, third paragraph, point (a) to read verbatim: 'Chapters I and II shall apply from 2 February 2025, with the exception of Article 5(1), first subparagraph, points (ba) and (bb), and Article 5(1a) and (1b) which shall apply from 2 December 2026;' This confirms the exact day (2 December 2026, not merely 'December 2026') for the Article 5(1) prohibitions on AI-generated non-consensual intimate imagery ('nudification') and CSAM (Prohibition 9), retracting the prior round's flag that only the month, not the day, was confirmed. Both quotes were read directly from the OJ PDF's own consecutively numbered amendment clauses (39) and (40) to Article 1 of Regulation (EU) 2026/1744, on page 35 of 41 of the official PDF, and independently corroborated via search results quoting the same amendment clauses verbatim.