If you did any AI Act planning before this summer, some of it is now out of date. The EU changed the deadlines in July 2026, and the change is largely good news for small businesses.
It is also the reason a lot of advice you will find is wrong. Anything written before late July carries a timetable that no longer applies.
General information, not legal advice. Dates here were checked against the official texts in August 2026.
Applies to you if
- You were preparing for high-risk obligations in August 2026, particularly around recruitment.
- You have a compliance plan, a consultant’s roadmap or a policy written before July 2026.
- You were told AI literacy meant guaranteeing every staff member reached a standard.
Probably does not apply if
- You had not started, in which case just use the current dates below.
- Your AI use is drafting and summarising with a person reviewing. The parts that moved were never going to reach you.
What happened
The Commission proposed a simplification package in November 2025, generally called the Digital Omnibus. The AI part became Regulation (EU) 2026/1744, which entered into force on 27 July 2026.
The pressure behind it was practical rather than political: the technical standards that high-risk compliance depends on were not ready, and industry pointed out that businesses cannot comply with a standard that does not exist yet.
What changed, and what did not
High-risk deadlines moved. Most Annex III high-risk obligations, including recruitment, education, credit and access to essential services, moved from 2 August 2026 to 2 December 2027. High-risk AI embedded in regulated products moved to 2 August 2028.
The delay became unconditional. Earlier drafts tied the new dates to standards being ready, which would have left everyone guessing. That trigger was removed. These are now fixed calendar dates that cannot move again without new legislation, which is arguably more useful than the extra time.
AI literacy was softened. Article 4 previously required measures to ensure a sufficient level of AI literacy. It now requires measures to support the development of AI literacy. An obligation of effort rather than of result: you are no longer expected to guarantee that every individual reaches a measurable standard.
Transparency did not move. The Article 50 duties started on 2 August 2026 as originally planned. If you were hoping labelling had been postponed, it was not.
Two prohibitions were added. From 2 December 2026, generating or manipulating non-consensual intimate material, and material depicting child sexual abuse, are prohibited practices. Not a small-business compliance question, but worth knowing the prohibited list grew rather than shrank.
Penalties did not change. The tiers are unchanged: up to €35 million or 7% of worldwide turnover at the top, with SMEs capped at the lower of the two figures.
The timetable as it now stands
- 2 February 2025. Definitions, prohibited practices, AI literacy. In force.
- 2 August 2025. Governance and general-purpose model obligations. In force.
- 2 August 2026. General application, enforcement, Article 50 transparency. In force.
- 2 December 2026. Transitional machine-readable marking deadline for existing generative systems, and the two new prohibitions.
- 2 December 2027. Main Annex III high-risk obligations.
- 2 August 2028. High-risk AI in regulated products.
Three situations, worked through
A firm that bought CV screening software and panicked in June. The obligations they were rushing to meet now land in December 2027. The right response is to use that time to find out what the tool actually does, not to forget about it.
A business quoted for a certified AI literacy programme. The wording softened, and it never named a certificate in the first place. A written policy, a briefing and a refresher is proportionate for a small team.
A practice that assumed labelling was postponed too. It was not. Transparency duties started on 2 August 2026, so a chatbot should say it is a chatbot now, not in 2027.
Five things worth doing
- Check the date on any AI Act advice you hold. Anything before late July 2026 has the old timetable.
- Move your high-risk work to a December 2027 plan rather than abandoning it. The date is now fixed.
- Do the transparency items now. Those did not move.
- Keep the literacy measures proportionate and write them down. Softer wording is not no wording.
- Re-read any quote for compliance services written against the old dates, and ask what has changed in it.
Common misconceptions
- “The AI Act was delayed.” Parts of it were. General application and enforcement began on 2 August 2026 as planned.
- “The dates could move again.” The conditional trigger was deliberately removed. Another change would need fresh legislation.
- “AI literacy was dropped.” Softened, not dropped, and supervision of it began in August 2026.
- “The EU is rolling back AI regulation.” The timetable moved and the prohibited list grew. That is sequencing, not retreat.
The honest summary
If you screen job applicants with software, you gained sixteen months. If you use AI to draft and summarise, nothing that moved was ever going to reach you, and the parts that do reach you started in August 2026.
The most useful thing to take from this is a habit rather than a date: check when AI Act advice was written before acting on it. The current position is set out here. If you have a plan built on the old timetable and want to know what still stands, describe what you were preparing for.
Sources
Regulation (EU) 2026/1744 and Regulation (EU) 2024/1689 on EUR-Lex, and the European Commission’s AI Act pages. Checked 6 August 2026.
Common questions
Was the whole AI Act delayed?
No. General application, enforcement and the Article 50 transparency duties began on 2 August 2026 as planned. What moved was the main high-risk obligations, to 2 December 2027, and product-embedded high-risk AI, to 2 August 2028.
Could the deadlines move again?
Not easily. Earlier drafts tied the new dates to European standards being ready. That conditional trigger was removed from the final text, so these are fixed calendar dates that would need fresh legislation to change.
Does the softened AI literacy wording mean we can ignore it?
No. It became an obligation of effort rather than of result, and market surveillance authorities began supervising it in August 2026. A written policy and a proportionate briefing is still the expectation.