Key Developments

The European Union has significantly advanced its AI safety framework with major updates to the AI Act implementation timeline. The Council of the EU has extended the deadline for establishing national AI regulatory sandboxes to December 2, 2027, while publishing the Second Draft Code of Practice on AI-generated content marking and labeling. Most remaining AI Act provisions are scheduled to activate on August 2, 2026.

Concurrently, the European Parliament adopted new copyright protection measures specifically targeting generative AI systems, mandating full transparency and fair remuneration when copyrighted materials are used for AI training.

Industry Context

These developments come as recent research reveals critical gaps in AI safety evaluation. A landmark study on “Intent Laundering” demonstrated that safety conclusions break down when triggering cues are removed, with jailbreaking techniques achieving 90-98% success rates. The 2026 International AI Safety Report, compiled by over 100 experts from 30+ countries, identified cybersecurity as a domain where real-world AI harm is now strongest.

The regulatory push gains urgency as research shows structural fragmentation in AI safety work, with only 5% of papers bridging the gap between safety and ethics communities.

Practical Implications

For Irish and European AI developers, these changes create both opportunities and compliance requirements. The extended sandbox timeline provides more preparation time for testing regulatory frameworks, while the August 2026 deadline for core provisions remains firm. Companies using AI for content generation must now implement transparent labeling systems and establish copyright compliance mechanisms.

The EU’s clarification of AI Office competences versus national authorities is particularly relevant for Irish fintech and border management applications, where national oversight remains paramount.

Open Questions

Key uncertainties remain around enforcement mechanisms for the new copyright requirements and how the “safety-security dilemma” identified in recent research will be addressed in practice. The effectiveness of regulatory sandboxes in bridging the safety-ethics divide also remains to be demonstrated, as does coordination between EU-level AI Office oversight and national regulatory approaches across member states.

Source: European Parliament