The agreement stands — and the list of prohibitions stays as it is
In May, the decision on postponing the AI obligations is taken. In the same month, the EU examines for the first time whether its list of prohibited AI practices needs changing. The answer: no.
After the failed round of negotiations on 28 April, it took nine days. In the night leading into 7 May 2026, the Council and Parliament agreed on the «Digital Omnibus on AI» — the postponement of the high-risk obligations of the AI Act.
The Council announced the agreement the same day: Council of the EU, press release of 7 May 2026. On 13 May, the Committee of Permanent Representatives confirmed the provisional text — an intermediate step, not a new decision.
A side note that shows how easily dates slip: some sources give 6 May. That is the day the negotiation began. Agreement was reached in the early hours of the 7th.
The month's second piece of news
Less noticed, but at least as important in practice: on 20 May 2026, the Commission presented its first report reviewing the prohibited AI practices, published on 22 May. It is the first report of its kind under Article 112 of the AI Act: Report COM(2026) 234 final on the review of the prohibitions.
The list of prohibitions in Article 5 names practices that are simply not permitted in the EU — such as rating people according to their social behaviour or deliberately exploiting vulnerabilities. It has applied since 2 February 2025 and was never affected by the postponement.
What is prohibited stays prohibited. What was postponed is the assessment obligation for permitted systems — not the list of prohibitions.
The outcome of the review
The majority of the member states involved saw no need for change. The reasoning is sober and understandable: the prohibitions have only applied since February 2025, enforcement is not yet running smoothly in many places, and there is simply a lack of practical experience.
One gap was named: AI used to generate abuse imagery and non-consensual intimate content. It was closed not through this report but through the Digital Omnibus.
What businesses take away from this
Anyone operating an AI application should know the difference between the two pieces of news. One postpones a deadline for systems that are permitted but subject to assessment. The other confirms a list of what is simply not on — and nothing about that has changed, nor will it for the time being.
Checking whether your own system falls under one of the prohibitions is therefore neither postponed nor settled. It has been due since February 2025.
Sources
This is how our system does it