Source: Edotto

On 7 May 2026 the Council of the European Union and the European Parliament reached a provisional agreement under the Omnibus VII package to amend the EU Regulation on artificial intelligence (AI Act). The agreement delays compliance deadlines for high-risk AI systems, distinguishing between independent high-risk systems (new application date 2 December 2027) and AI systems integrated into sectoral products (new application date 2 August 2028), and postpones the deadline for member states to establish AI regulatory sandboxes to 2 August 2027. It also shortens the tolerance period for implementing required measures from six to three months, setting a new cutoff of 2 December 2026 for that transitional tolerance.

The colegislators reintroduced an obligation to register AI systems in the EU database even for providers who consider their systems not high-risk, and strengthened transparency duties requiring providers to clearly identify AI-generated content. The pact adds an explicit prohibition on using AI to generate non-consensual sexual deepfakes and artificially created sexual-abuse content involving minors, increases provider responsibilities to prevent such content and protect minors, and restores the “strict necessity” criterion for processing special categories of personal data for bias detection and mitigation. The agreement also enhances coordination between the AI Act and sectoral EU rules (medical devices, machinery, toys, elevators, personal watercraft), expands competences of the European AI Office over general-purpose models developed by a provider, and foresees implementation acts and guidance from the European Commission to limit overlap with sectoral obligations; final formal approval and legal-linguistic revision by the colegislators remain pending.