EU AI Act
EU's first comprehensive AI risk-tier law; GPAI enforcement begins 2 August 2026.
The Commission's own guidance sets the Code of Practice signatory deadline at 27 July 2026, five days before Article 50 becomes legally binding on 2 August; a correction issued on 25 July fixed an earlier report that had wrongly given 22 July.
Last refreshed: 4 August 2026 · Appears in 3 active topics
Which AI systems still face the 2 August 2026 Article 50 deadline?
Timeline for EU AI Act
AI Office gains enforcement powers in August
European Tech SovereigntyAI Act fines land on a stale register
European Tech SovereigntyMentioned in: Meta tells the SEC what broadcasters do not
Media's AI PivotOmnibus widens the AI Office's reach
European Tech SovereigntyMentioned in: Correction: the EU deadline is 27 July
Media's AI PivotBackground
The EU AI Act is the world's first comprehensive legal framework for artificial intelligence, passed by the European Parliament in March 2024 and entering staged enforcement from August 2025. It classifies AI systems into four risk tiers: unacceptable risk (prohibited outright), high risk (mandatory conformity assessments), limited risk (transparency obligations) and minimal risk (no requirements).
High-risk categories include AI used in employment decisions, education, credit scoring, law enforcement and critical infrastructure. The Act also imposes distinct obligations on providers of general-purpose AI (GPAI) models, chiefly OpenAI, Google DeepMind and Anthropic, including technical documentation, capability evaluations and incident reporting, backed by fines of up to 7% of global annual turnover for the most serious violations .
The Act has become a central battleground in Europe's technology-sovereignty debate: supporters call it a global gold standard giving European regulators real leverage over US tech firms, while critics, including Mistral CEO Arthur Mensch, warn it risks handicapping European AI labs against US and Chinese rivals facing no equivalent burden. South Korea's rival AI Basic Act, in force since January 2026, took the opposite innovation-first approach, sharpening the contrast .
Its content-labelling deadline keeps shifting
The Commission finalised its AI content-marking Code of Practice on 10 June 2026, requiring AI-generated content to be labelled from 2 August, with systems already deployed given until 2 December 2026 to comply . Zero EU broadcasters or media companies had signed at finalisation, and none had by 15 July either, when reporting gave 22 July as the signatory-list cutoff.
That date itself proved wrong: a correction on 25 July fixed the Commission's own guidance page at 27 July, 18:00 CEST, just five days before Article 50 becomes legally binding . Even the law's own Deadline-tracking, in other words, cannot yet keep pace with the law itself.
Employment AI rules slipped to 2027
A provisional Digital Omnibus deal reached in the early hours of 7 May 2026 dropped the binding employer AI-literacy duty entirely, replacing it with a non-enforceable instruction for member states to "encourage" AI literacy, and pushed high-risk employment AI obligations under Annex III from August 2026 to December 2027 .
The Council gave the package final approval on 29 June, confirming employers must only "support" staff understanding of workplace AI rather than guarantee it . The 16-month delay landed as AI-attributed job cuts kept climbing, leaving workers subject to AI-driven hiring screens without EU-level recourse for well over a year.
One government order silenced two models
On 12 June 2026 a US directive ordered Anthropic to block foreign nationals, including EU institutions, from its Fable 5 and Mythos 5 models; unable to screen nationality mid-conversation, Anthropic switched both off for every customer on earth . The EU's own cybersecurity agency, ENISA, had joined Anthropic's restricted access programme in April and lost it within weeks, prompting an MEP to call the episode a "kill-switch".
The timing cuts directly against the Act's own framework: GPAI enforcement, meant to give Brussels leverage over frontier providers, activates on 2 August, yet the episode showed Washington can revoke access to those same providers' models unilaterally. The partial US reversal on 27 June, five weeks before that Deadline, did not remove the underlying precedent .
Its content-labelling deadline keeps shifting
What is the EU AI Act's Article 50 and when does it apply?
What is the difference between the GPAI and high-risk AI deadlines in the EU AI Act?
What did the EU AI Act Digital Omnibus change in May 2026?
What happens when the DMA Google decision and AI Act enforcement collide?
What fines can the EU AI Act impose on OpenAI?
When does the EU AI Act take full effect?
Does the EU AI Act hurt European AI companies?
When must broadcasters sign the EU AI Act's content-marking Code of Practice?
Has the EU delayed its AI workplace rules?
When does the EU start enforcing rules on ChatGPT and other AI models?
