Blog / Regulation
The AI Act after the Omnibus: what applies to AI agents, date by date
Published in the Official Journal on 24 July 2026, the Digital Omnibus on AI moves some AI Act deadlines and leaves the rest untouched. For a company running agents, the question is no longer whether the Act applies but which of its decisions it targets, and from when.
By Vincent Dorange, , 4 min read
high-risk rules start to apply to the sensitive areas listed in Annex III
On 24 July 2026 the Official Journal of the European Union published Regulation (EU) 2026/1744, which everyone calls the Digital Omnibus on AI 1. Adopted on 8 July and in force since 27 July, it amends the AI Act a few days before a deadline many companies dreaded, 2 August 2026. Its effect is hard to sum up in a sentence. Part of the obligations arrives on time, another part is pushed back by 16 months or 2 years, and the text still says nothing about agents even though they now raise the most concrete questions.
What applies on 2 August
The AI Act applies in stages 2. Prohibited practices and the AI literacy duty have applied since 2 February 2025, the rules on general-purpose models and the governance provisions since 2 August 2025. 2 August 2026 marks the general application of the Act, including the transparency duties of Article 50: a system intended to interact directly with people must be designed so that they know they are dealing with an AI, unless this is obvious to a reasonably well-informed person.
For an agent that answers customers, negotiates a delivery date or chases a supplier by email, that duty therefore applies within days, and the Omnibus does not postpone it. It touches it on one point only, the machine-readable marking of generated content (Article 50, paragraph 2): systems placed on the market before 2 August 2026 have until 2 December 2026 to comply 1.
What the Omnibus pushes back
The obligations for high-risk systems, those that require risk management, technical documentation, logging and human oversight, are postponed. For the sensitive areas of Annex III (employment, education, access to credit, life and health insurance pricing, essential public services, among others) the date is now 2 December 2027. For systems built into products already regulated under Annex I it is now 2 August 2028 1.
The same regulation adds 2 prohibitions to Article 5, applicable from 2 December 2026: systems that generate realistic intimate images of identifiable people without their consent, and systems that produce child sexual abuse material. It rewrites Article 4 on AI literacy, which now asks providers and deployers to take measures to support the development of their staff's AI literacy without requiring them to guarantee any particular level. It also extends to small mid-cap companies several simplifications that were reserved for SMEs, such as simplified technical documentation 1.
Articles 12 (record-keeping), 14 (human oversight), 19 (automatically generated logs) and 26 (obligations of deployers) are not amended 1. Their content is the 2024 text; only the date from which they apply has changed.
An Act that never says "agent"
The English text of the 2024 Act does not use the word "agent" 2, and the Omnibus did not change that. An agent is neither more nor less regulated than any other AI system: it is classified by what it does. A customer service agent falls under Article 50. The same agent, if it assesses a person's creditworthiness or prices a health insurance policy, enters Annex III and will become high-risk on 2 December 2027. Everything depends on the decisions it is given, which is exactly the angle taken by the Agentic Commerce Framework®.
The GDPR, for its part, postponed nothing. Its Article 22 gives everyone the right not to be subject to a decision based solely on automated processing that produces legal effects concerning them or similarly significantly affects them, unless the decision is necessary for a contract, authorised by law or based on explicit consent 3. An agent that refuses a refund or rejects a job application without human involvement runs into that text today.
What I recommend doing before December 2027
The delay buys time if it is used to know one's agents decision by decision. Three workstreams, in this order, are enough to turn these 16 months into a head start rather than a reprieve.
- Keep an AI inventory of the agents in service, including those a team plugged in without approval, then record for each one the decisions it actually takes, which often go beyond its product sheet.
- Sort those decisions by the text that targets them: Article 50 only, Annex III, or GDPR Article 22. The ACF criticality score, computed as (impact × frequency × irreversibility) ÷ 10, helps order the work by starting with what would cost most to put right.
- Put in place now what Articles 12 and 14 will require: an audit trail that shows who authorised what, and human oversight placed where the decision truly commits the organisation.
A company that does this work before 2027 will reach the deadline with evidence to show. One that waits for the deadline will discover its agents on the day an inspector asks it to describe them.
Sources
Links to the official texts and original publications.
- 1Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), Official Journal of the European Union, L series, 24 July 2026.
- 2Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence, Official Journal of the European Union, L series, 12 July 2024.
- 3Regulation (EU) 2016/679 (GDPR), Article 22, Official Journal of the European Union, 27 April 2016.
Glossary terms used in this article
Vincent Dorange, creator of the Agentic Commerce Framework®