First AI Act harmonised standard hits the Official Journal; Digital Omnibus is law; Article 50 FAQ sets a 2 December date

Article 50 — Applies to providers and deployers since 2 August 2026: chatbot disclosure, machine-readable marking of synthetic output, and deepfake labelling (AI Act Article 50). AESIA (Spain)

2 December 2026 — AI systems marketed before 2 August 2026 must meet the Article 50 marking obligations by 2 December 2026, per Spain's AESIA FAQ on the Commission transparency guidelines. AESIA (Spain)

2 August 2027 — The date by which every remaining AI Act provision applies, including the high-risk requirements, under Regulation (EU) 2024/1689 as amended by the Omnibus. EUR-Lex

AESIA FAQ on transparency — The Spanish AI supervisory agency published a question-and-answer page on the Commission's guidelines for AI-generated content, covering who counts as a provider or deployer, the four cumulative criteria for when a chatbot must disclose itself, the three-part deepfake test in Article 3(60), and which outputs — source code, machine-to-machine results, closed industrial settings — are exempt from marking. It is a ready-made briefing for non-legal teams facing the transparency duties, and it states the fine exposure plainly. Penalties run up to 15 million euros or 3% of worldwide turnover for the previous financial year, and adherence to the voluntary Code of Practice on Transparency is offered as a route to demonstrate compliance. AESIA (Spain)

Regulation (EU) 2026/1744 — Adopted on 8 July 2026 and in force, it amends the AI Act along with the aviation and machinery regulations to ease implementation. The 'safety component' definition is narrowed so fewer systems count as high-risk; the Article 4 duty is softened from ensuring staff AI literacy to supporting its development; and the legal basis for processing special-category data to detect and correct bias now reaches deployers and non-high-risk systems, under Article 10(5)-style safeguards. High-risk classifications made under the old safety-component wording may need a second look, and literacy programmes planned against the harder Article 4 reading can be recalibrated. The recitals point to late standards and slow national governance set-up as the reason the compliance load came out heavier than expected. EUR-Lex

The EU AI Act Newsletter #112 — The Future of Life Institute's 6 October issue reports that tech Commissioner Henna Virkkunen, speaking at the RAID Conference, pledged to keep pursuing an international AI security agreement despite US opposition, with the EU backing a Finland- and Norway-led initiative for an AI safety monitoring body signed by 20 further countries. MEP Axel Voss publicly said the Commission is too slow on AI and that enforcement of the Act falls short of what he wants, pointing at the AI Office. These are on-the-record signals about political appetite for enforcement now that the AI Office has been in charge of general-purpose model oversight since August. The issue also notes the Commission designated ChatGPT a Very Large Online Search Engine under the DSA, prompting analysis of how duties divide between that law and the AI Act. The EU AI Act Newsletter (Future of Life Institute)