EU Postpones High-Risk AI Act Deadlines to 2027/2028, But Strict Watermarking Rules Take Effect August 2, 2026

1. Introduction: The EU AI Act Delay

The European Parliament has approved amendments to the landmark **EU AI Act** via the "Digital Omnibus" package. The new amendments provide much-needed relief to enterprise developers by pushing back compliance deadlines for high-risk artificial intelligence systems. However, this delay does not extend to transparency obligations: strict labeling and digital watermarking requirements for all AI-generated content are still set to become legally binding in just one month.

As governments struggle to balance rapid innovation with safe governance, corporate AI compliance has become a central focus. For reference on the scale of industrial compute and national security reviews, read about the Google Gemini 2.5 Pro reasoning mode and delayed releases.

2. Postponed Deadlines for High-Risk AI Systems

Under the approved amendments, the deadlines for high-risk AI systems (defined as AI used in critical infrastructure, education, employment, and law enforcement) have been pushed back:

• Standalone High-Risk Systems: Pushed back from 2026 to December 2, 2027.
• Embedded High-Risk Systems: (AI integrated into physical products like medical devices or vehicles) Pushed back to August 2, 2028.

European regulators noted that the delay is meant to give businesses and developers sufficient time to adapt their systems and prevent market disruption.

3. The Imminent August 2, 2026 Transparency & Watermarking Mandate

While high-risk systems have received a delay, Article 50 transparency obligations **will not be postponed**. Starting **August 2, 2026**, all AI-generated media (text, audio, video, and images) must be clearly labeled to prevent public deception and combat deepfakes. This includes:

• Metadata Watermarking: Embedding cryptographically secure, machine-readable watermarks indicating the media is AI-generated.
• User Disclosure: Clearly notifying users whenever they are interacting with an automated AI agent or system.

Failing to comply with these watermarking requirements could lead to significant fines. Creators and businesses using AI generation tools must start implementing metadata embedding immediately. For details on how AI capabilities are transitioning to background execution, see our report on Anthropic's Economic Index and the shift toward autonomous agents.

4. EU Commission's Voluntary Code of Practice on Transparency

To help bridge the gap before the August mandate, the European Commission has published a voluntary **Code of Practice on Transparency of AI-Generated Content**. The code outlines technical standards for watermarking, including support for open-source metadata standards (like C2PA). Major players in the image generation and LLM fields have already pledged support to ensure their outputs include compliant metadata by August.

5. The Global Regulatory Domino Effect (US & California Updates)

The EU’s actions are accelerating regulatory moves worldwide. In the United States, bipartisan senators recently released a draft of the **"Great American AI Act"**, which focuses on security standards, national compute networks, and international trade restrictions. Simultaneously, California has advanced state-level bills requiring employers to notify employees if automated systems are used for hiring decisions or task distribution. This focus on regulatory and hardware security mirrors the BIS warning regarding the $1 trillion AI credit bubble.

6. Frequently Asked Questions (FAQ)