19 June 2026

Legal Update: Postponements and Adjustments to the European Union’s AI Legislation

The European Parliament has adopted amendments to several key provisions of the EU AI Act as part of the Digital Omnibus legislative package.

Revised implementation timetable

The application of the rules governing stand-alone high-risk AI systems has been postponed until 2 December 2027.
For AI systems incorporated into products as safety components and governed by sector-specific legislation, the compliance deadline has been extended to 2 August 2028.
The obligation to affix a watermark to AI-generated content has been postponed until 2 December 2026. Until then, a machine-readable label will remain temporarily sufficient.

New prohibited practices

The creation, development and dissemination of child sexual abuse material using AI technologies are prohibited.
AI systems used for “nudification”, or to generate sexual images or audio content depicting a person without their explicit consent, are also banned.
Both providers and deployers may be held liable, unless technical safeguards preventing such uses are implemented by 2 December 2026.

Simplification and the removal of overlapping regulatory requirements

The proposal also seeks to simplify administrative requirements and eliminate instances of overlapping regulation.
AI-related requirements applicable to machinery and related products will be harmonised in order to avoid duplicate compliance obligations, with reliance placed on the existing sector-specific product safety framework.
AI systems used solely to optimise performance will no longer be classified as high-risk where their failure does not directly threaten health or safety.
The processing of personal data will now be permitted, subject to strict safeguards, where necessary to identify and correct discriminatory bias in AI systems.
Regulatory relief initially reserved for small and medium-sized enterprises will be extended to small mid-cap companies in order to support their continued growth.
The implementation of the rules applicable to general-purpose AI systems will also be streamlined and placed under the central supervision of the EU AI Office.

These provisions must still be adopted by the Council of the European Union before they can enter into force.

FIRSH Expert Insight

New restrictions have been introduced to regulate the use of certain AI tools, particularly in response to sexually explicit deepfakes.

This issue has been central to the work of FIRSH’s research laboratory for the past three years and was examined extensively in our White Paper. Several of the newly proposed measures reflect recommendations made in that publication.

The relaxation of the legislative timetable should not be interpreted as a signal that compliance efforts may be reduced. Rather, it creates a strategic opportunity for organisations to secure their compliance framework without unnecessary haste.

FIRSH assists businesses in mapping their AI systems, auditing their providers and requiring the implementation of content-marking mechanisms from the outset.