Key Takeaways
- • High-risk AI compliance deadlines extended by 18 months to 2 December 2027 for Article 6(2) systems and two years to 2 August 2028 for Article 6(1) systems, deferring obligations originally scheduled for 2 August 2026 due to delayed standards development and competent authority establishment that would have increased implementation costs prohibitively.
- • Non-consensual intimate imagery AI systems now prohibited as of 2 December 2026, expanding the EU's restricted AI list to address technological and societal developments including deepfake synthetic media and child sexual abuse material generation.
- • Transparency requirements deadline maintained at 2 August 2026 for most AI systems, with a four-month extension to 2 December 2026 solely for synthetic audio, image, video, and text-generating systems, preserving near-term disclosure obligations while acknowledging sector-specific implementation challenges.
- • AI literacy obligations softened from mandatory "ensure sufficient level" to "support development" standard, reducing compliance burden on providers and deployers while shifting responsibility to the European Commission and Member States to provide training and resources.
- • Supply chain penalties clarified at up to 3% of global annual turnover or €15 million (~$17 million), whichever is higher, for breaches of original provider cooperation obligations with new providers regarding information sharing and technical access.
Summary
The EU AI Act Amendment, published in the Official Journal on 24 July 2026 and effective as of 27 July 2026, introduces material compliance timeline adjustments and regulatory clarifications following negotiations between the European Parliament, Council, and Commission. The amendment addresses implementation challenges identified during the first wave of AI Act rollout, specifically acknowledging delays in harmonized standards development, common specifications, and national competent authority establishment that threatened the original 2 August 2026 deadline for high-risk AI system obligations.
The amendment provides differentiated compliance extensions depending on regulatory classification. High-risk AI systems classified under Article 6(2) and Annex III receive an 18-month extension to 2 December 2027, while those under Article 6(1) and Annex I—which reference specific EU product legislation—receive a two-year extension to 2 August 2028. This tiered approach reflects varying complexity levels across different high-risk AI applications. The transparency requirements baseline remains at 2 August 2026 for most systems, but synthetic content generators (audio, image, video, and text) receive a four-month grace period to 2 December 2026, signaling regulatory acknowledgment of distinct implementation challenges in generative AI disclosure standards.
New prohibitions targeting non-consensual intimate imagery represent a substantive expansion of restricted AI use cases. AI systems that generate or manipulate non-consensual intimate material, including deepfakes and related synthetic media, become prohibited as of 2 December 2026. This addition reflects documented technological and societal developments since the original AI Act adoption, including widespread deployment of generative models capable of creating realistic manipulated content.
The amendment softens AI literacy obligations while introducing supporting infrastructure. Rather than requiring providers and deployers to "ensure, to their best extent, a sufficient level" of staff AI literacy, the revised standard requires them to "support the development" of such capabilities—a materially lower compliance bar. Simultaneously, the amendment obligates the European Commission and Member States to facilitate compliance through training programs, informational resources, and best-practice exchanges, recognizing that the previous framework created disproportionate burdens across heterogeneous operator categories.
Supply chain responsibilities are now formalized with explicit penalties. The amendment details original provider obligations to cooperate with subsequent providers regarding information disclosure and technical access, with breaches triggering fines up to 3% of global annual turnover or €15 million (~$17 million), whichever is higher. Additional technical amendments clarify the "safety component" definition for high-risk systems, strengthen AI Office market surveillance authority over general-purpose AI and very large online platforms, and align AI Act requirements with the Cyber Resilience Act, GDPR, and other EU product legislation. The European standardisation organisations published EN 18286 on 22 July 2026 and committed to publishing remaining high-risk standards in late 2026 and early 2027.
The amendment provides differentiated compliance extensions depending on regulatory classification. High-risk AI systems classified under Article 6(2) and Annex III receive an 18-month extension to 2 December 2027, while those under Article 6(1) and Annex I—which reference specific EU product legislation—receive a two-year extension to 2 August 2028. This tiered approach reflects varying complexity levels across different high-risk AI applications. The transparency requirements baseline remains at 2 August 2026 for most systems, but synthetic content generators (audio, image, video, and text) receive a four-month grace period to 2 December 2026, signaling regulatory acknowledgment of distinct implementation challenges in generative AI disclosure standards.
New prohibitions targeting non-consensual intimate imagery represent a substantive expansion of restricted AI use cases. AI systems that generate or manipulate non-consensual intimate material, including deepfakes and related synthetic media, become prohibited as of 2 December 2026. This addition reflects documented technological and societal developments since the original AI Act adoption, including widespread deployment of generative models capable of creating realistic manipulated content.
The amendment softens AI literacy obligations while introducing supporting infrastructure. Rather than requiring providers and deployers to "ensure, to their best extent, a sufficient level" of staff AI literacy, the revised standard requires them to "support the development" of such capabilities—a materially lower compliance bar. Simultaneously, the amendment obligates the European Commission and Member States to facilitate compliance through training programs, informational resources, and best-practice exchanges, recognizing that the previous framework created disproportionate burdens across heterogeneous operator categories.
Supply chain responsibilities are now formalized with explicit penalties. The amendment details original provider obligations to cooperate with subsequent providers regarding information disclosure and technical access, with breaches triggering fines up to 3% of global annual turnover or €15 million (~$17 million), whichever is higher. Additional technical amendments clarify the "safety component" definition for high-risk systems, strengthen AI Office market surveillance authority over general-purpose AI and very large online platforms, and align AI Act requirements with the Cyber Resilience Act, GDPR, and other EU product legislation. The European standardisation organisations published EN 18286 on 22 July 2026 and committed to publishing remaining high-risk standards in late 2026 and early 2027.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.