EU AI Act: Transparency Rules Launch Aug 2026, High-Risk Delayed

The European Union's AI Act, a significant piece of legislation, is entering a new phase of applicability. While many anticipated a broad rollout of its core obligations, recent amendments have created a two-speed timetable. On August 2, 2026, specific transparency requirements will take effect, but the more stringent high-risk AI regulations have been postponed Source.

This shift, formalized by the Digital Omnibus on AI signed on July 8, 2026, means much of the previous guidance on the AI Act's 2026 timeline is now outdated. Business leaders should understand what is genuinely coming into force and what has been deferred.

What Becomes Applicable in August 2026

The crucial element becoming enforceable on August 2, 2026, is Article 50, which mandates general transparency. This applies to any AI system introduced into the EU market and covers three main areas:

• Chatbot Disclosure: Systems designed to interact directly with people must clearly indicate that the user is engaging with an AI, unless this is inherently obvious from the context.
• Synthetic Content Labeling: Artificially generated content, including deepfakes and manipulated audio, image, or video, must be disclosed as such. This also includes a machine-readable marking obligation for synthetic audio, image, video, and text Source. Providers whose systems are already on the market by August 2, 2026, have until December 2, 2026, to implement these markings.
• Biometric System Information: Emotion recognition and biometric categorization systems must inform individuals who are subject to their use.

Other earlier provisions of the AI Act, such as prohibitions on certain AI practices (in effect since February 2025) and rules for general-purpose AI (since August 2025), remain unchanged.

Deferral of High-Risk AI Obligations

The Digital Omnibus on AI has significantly altered the timeline for high-risk AI systems, extending compliance deadlines in two stages:

• Stand-alone Annex III Systems: AI systems listed in Annex III, such as recruitment tools, credit scoring, educational applications, law enforcement, border control, and critical infrastructure, now have until December 2, 2027, for full compliance, an extension of 17 months.
• Embedded AI in Regulated Products (Annex I): AI systems integrated into products already covered by existing EU product safety laws (e.g., medical devices, machinery, toys) will not face full compliance until August 2, 2028.

The reason for these deferrals was primarily procedural. Member states were slow in appointing national competent authorities, and the necessary harmonized standards and conformity assessment tools, critical for high-risk compliance, were not yet finalized Source. It is important to note that these delays provide more time, but they do not soften the underlying obligations for high-risk AI.

Unanticipated Additions and Institutional Shifts

Beyond just rescheduling, the Digital Omnibus introduced a few unexpected changes:

New Prohibited AI Practice

Article 5 now includes a prohibition on AI systems designed to generate non-consensual intimate imagery, often referred to as 'nudifier' applications. This new ban applies from December 2, 2026 Source.

Expanded Powers for the AI Office

The AI Office, based in Brussels, has gained broader supervisory authority. Its reach now extends beyond general-purpose AI models to the systems built upon them, particularly for vertically integrated providers where the model and system originate from the same business group. This centralizes oversight for many frontier AI developers.

Regulatory Sandboxes Delayed

Member states are now required to establish at least one national regulatory sandbox by August 2, 2027, a year later than initially planned. An EU-level sandbox for continental-scale testing has also become an option.

Business Implications

For businesses deploying AI tools that interact with customers or generate content for the EU market, such as chatbots or synthetic media, the August 2026 deadline is firm. Compliance will require immediate attention to disclosure and labeling practices.

For those developing high-risk AI, the delays offer a crucial reprieve to prepare. However, the consistent message is that the EU is committed to regulating AI. The Act's first amendment within two years of adoption, even before its central provisions applied, demonstrates an evolving regulatory landscape that businesses must continuously track and adapt to. Future compliance will depend heavily on the finalization of harmonized standards, which, if delayed again, could lead to further adjustments.