EU delays high-risk AI rules, doubles down on transparency

The gist
The EU has hit pause on strict high-risk AI rules, but is doubling down on transparency—so your AI must still come clean, even if the toughest compliance hurdles are delayed.
What to know
- The EU’s Digital Omnibus has postponed most high-risk AI compliance deadlines from August 2026 to December 2027 (and some to August 2028), giving companies extra breathing room to adapt.
- Transparency rules start August 2, 2026, requiring clear disclosure of AI use in user interactions, AI-generated content, emotion recognition, and deepfakes—with penalties up to €35 million or 7% of global turnover for non-compliance.
- The European Commission’s upcoming guidelines and a voluntary AI transparency code offer practical tools like robust watermarking, letting companies comply without obvious labels while gaining legal cover.
Deadline Extension, New Uncertainty
The EU’s delay of high-risk AI compliance buys businesses time but leaves legal teams navigating a shifting regulatory landscape without clear guidance.
The EU's Digital Omnibus on AI has decisively postponed the compliance deadlines for high-risk AI systems, shifting the original enforcement date from August 2, 2026, to December 2, 2027, with some obligations extending even further to August 2028. This extension acknowledges the complexity of the AI Act's requirements and grants businesses vital additional time to develop the necessary governance structures, clarify supplier roles, and implement human oversight mechanisms. By easing the timetable, the EU aims to provide companies with a more realistic window to prepare for the demanding regulatory landscape surrounding high-risk AI technologies.
While the postponement alleviates immediate pressure on organizations deploying high-risk AI systems, it simultaneously introduces a layer of last-minute regulatory uncertainty for businesses and legal professionals striving to align with the evolving framework. As highlighted by French analysis noting "entreprises et juristes dans une incertitude de dernière minute," stakeholders must navigate this shifting timeline carefully, balancing the relief of extended deadlines with the ongoing need to comply with transparency obligations that remain effective from August 2, 2026. These include critical rules mandating disclosure when users interact with AI or when content is AI-generated, underscoring that not all regulatory requirements have been deferred.
The rationale behind this deferral is rooted in the EU's recognition that the technical standards, regulatory infrastructure, and compliance tools essential for implementing the AI Act's high-risk provisions are not yet sufficiently mature. As noted by TechTarget, the lack of definitive guidance has created practical challenges for enterprises expected to comply without a clear roadmap. This postponement thus reflects a pragmatic approach, allowing time for the development of robust frameworks that ensure proportionality, enhance fundamental rights protections, and bolster legal certainty within the AI ecosystem.
Transparency Rules With Global Reach
Mandatory AI transparency disclosures—enforced with steep fines—will apply to EU and non-EU companies alike, with machine-readable markings replacing visible labels as the new compliance standard.
Effective August 2, 2026, the EU AI Act enforces comprehensive transparency obligations requiring both providers and deployers of AI systems to disclose AI involvement across four critical domains: direct user interactions, AI-generated content, emotion recognition and biometric categorisation, and deepfakes including AI-generated public-interest texts. Providers must inform users when they engage with AI systems unless it is 'almost no doubt left' to the average person, and ensure AI-generated outputs are marked in a machine-readable format, enabling detection without necessarily displaying visible labels. This framework extends extraterritorially, impacting non-EU companies like US firms serving EU customers, underscoring the EU’s broad regulatory reach.
To facilitate compliance and legal certainty, the European Commission issued detailed guidelines on July 20, 2026, clarifying Article 50’s scope and application, while the AI Office developed a voluntary Code of Practice on Transparency of AI-Generated Content. Organizations adopting this Code gain a presumption of conformity, benefiting from practical technical measures such as robust watermarking designed to survive content alterations, thereby enabling reliable traceability without overt user-facing labels. This dual approach balances enforceability with operational flexibility, encouraging proactive industry alignment.
Non-compliance with these transparency mandates carries steep penalties, with fines reaching up to €15 million or 3% of global annual turnover, and in some analyses, even up to €35 million or 7% of worldwide revenue, signaling regulators’ serious commitment to enforcement. This creates immediate pressure, especially on smaller AI providers lacking formal governance structures, to embed ethical oversight and transparency into AI deployment strategies. As Jan Karstens warns, the August 2 deadline is non-negotiable for disclosure, despite postponed deadlines for other high-risk AI obligations.
The transparency rules specifically target deceptive risks by mandating clear disclosure of deepfakes—particularly those involving identifiable persons in sensitive contexts—with audio deepfakes requiring plain-language disclaimers at the outset. However, content that has undergone human editorial review is exempt from disclosure, reflecting a nuanced approach to balancing transparency with editorial responsibility. Practical examples, such as labeling AI-generated training videos with EU-provided icons, illustrate the Act’s real-world application, reinforcing trust and credibility in AI use across sectors.
Balancing Rights and Innovation
The phased rollout strengthens fundamental rights protections and reduces compliance burdens for smaller firms, signaling a move toward proportional, innovation-friendly AI governance.
The Digital Omnibus on AI strategically enhances legal certainty by clarifying that the postponement of compliance deadlines for high-risk AI systems does not suspend existing obligations, compelling companies to maintain robust governance frameworks, define supplier responsibilities, and ensure human oversight and user training despite delayed deadlines. This phased implementation, with AI literacy requirements commencing February 2025 and transparency rules strengthening by August 2026, exemplifies a balanced approach that aligns regulatory demands with practical readiness, thereby reinforcing proportionality in the AI Act’s enforcement.
Fundamental rights protections receive a significant boost through reinforced transparency mandates under Article 50, requiring users to be clearly informed when interacting with AI systems and ensuring AI-generated content—whether audio, images, video, or text—is unmistakably labeled, including via machine-readable markings. Complementing these measures, the amendments introduce expanded allowances for fundamental rights impact assessments, particularly under Article 27, thereby deepening safeguards against potential rights infringements within AI deployments and reflecting the EU’s commitment to ethical AI governance.
Reflecting a nuanced and proportional regulatory philosophy, the Digital Omnibus eases certain burdens by softening AI literacy requirements and offering relief to small 'mid-cap' enterprises, which often face disproportionate compliance costs. Additionally, by reducing overlaps between the AI Act and existing EU product safety laws, the amendments alleviate the 'double burden' on businesses, facilitating smoother implementation and fostering a more innovation-friendly environment without compromising fundamental rights or legal clarity.
