August 2, 2026: What Changes Today for Organizations Using High-Risk AI
Today, August 2, 2026, the schedule for the EU AI Act takes a new turn. Omnibus Regulation (EU) 2026/1744, published in the Official Journal of the European Union on July 24, 2026, postpones the obligations for high-risk AI systems listed in Annex III until December 2, 2027, and those listed in Annex I until August 2, 2028.
But not everything is postponed. The transparency requirements under Article 50—notifying users when they interact with an AI system and labeling synthetic content—take effect today for all organizations, regardless of whether they have high-risk systems. The prohibitions, AI literacy requirements (effective Feb. 2025), and the penalty regime for GPAI models (effective Aug. 2025) remain fully in effect.
If your organization deploys chatbots or generates AI-generated content in the European Union, this date applies to you as well.
EU Regulation 2024/1689 · Art. 50 effective as of today · High-risk classification postponed (EU Omnibus Regulation 2026/1744)What does “high risk” mean under the EU AI Act
The practical definition is broader than many organizations expect, and the Omnibus Regulation provides more time to prepare—not a reason to stop doing so.
Resume screening, candidate ranking, performance evaluation
Credit scoring, insurance underwriting, loan decisions
Systems that determine eligibility for services or assistance
Energy, water, transportation, digital infrastructure
These requirements (Annex III) are no longer in effect as of today, but rather as of December 2, 2027. If you have any doubts about whether your system falls into this category, assess it now anyway—17 months is plenty of time to do it right, not to put it off.
What does the " EU AI Act" require, and when did it take effect?
Effective today, August 2, 2026 (Art. 50, Transparency), for any organization that uses generative or conversational AI:
AI Interaction Alert Today
If a chatbot or conversational system interacts with people, users must be informed that they are speaking with an AI, unless it is evident from the context.
Labeling Synthetic Content Today
Text, images, audio, or video generated or manipulated by AI must be labeled as such in a machine-readable format. Systems already on the market as of today have until December 2, 2026, to comply.
Postponed by the Omnibus Regulation; applicable to high-risk systems as of December 2, 2027 (Annex III) or August 2, 2028 (Annex I):
Technical Documentation Postponed
Description of the system, training data, performance, limitations, and controls in place. This information must be available for submission to the competent authority within the timeframe specified by the authority.
Event Registration Postponed
Automatic logs with sufficient granularity to identify risky situations and reconstruct decisions made by the system. Not generic logs, but decision traceability.
Documented human supervision Postponed
It is not enough for monitoring to be possible in theory. It must be documented that it took place—for which outputs, by whom, and with what results.
Active Risk Management System — Postponed
Not a one-time impact assessment. A continuous process of identifying, assessing, and mitigating risks— withevidence that it works.
Sanctions
The penalty regime (Art. 99) has been in effect since August 2025 and is not affected by the Omnibus Regulation. It already applies to violations of the transparency requirements under Art. 50, and will apply to high-risk violations as soon as those obligations become enforceable (2027/2028).
The percentage is applied to total revenue—not domestic revenue, nor revenue from the affected segment. Total revenue.
EU Regulation 2024/1689 · Article 99The problem most organizations face today
Postponing the high-risk designation does not eliminate the problem; it merely delays it. Organizations that have spent months preparing for compliance are discovering the same thing they always have: the documentation and records generated by their current systems do not meet the Regulation’s requirements for evidence.
A log in the platform’s own database does not constitute an event log with the level of integrity required by the EU AI Act. A human oversight policy signed by the steering committee does not prove that such oversight occurred for every relevant decision. And today, disclosure of interactions with AI and the labeling of synthetic content are already mandatory—there is no leeway there.
The EU AI Act does not require documentation. It requires verifiable evidence that controls were in place—currently for transparency, and starting in 2027 for high-risk cases.
The evidence required by the EU AI Act, starting today
V-PROOF It captures the applicable regulatory context, the AI-to-human ratio, authorship, and approval, and generates the V-Seal: SHA-256 + timestamp from the Base L2 blockchain.
The result: an event log with cryptographic integrity that can be independently verified, which no party can retroactively modify—useful for Article 50 today and prepared for the high-risk scenario when it arrives in 2027/2028.
The high risk has been postponed. Transparency, however, has not. The evidence you generate starting today will be available when you need it; evidence you do not generate cannot be created retroactively.
The transparency requirement under Article 50 is in effect. Can you prove that your controls were operational today?
Technical Documentation · Event Log · Verifiable Human Oversight.
Cryptographic evidence starting with the first asset, also ready for the high-risk requirements of 2027.
● Take action now
- Transparency under Article 50Indicate that the conversation is with an AI and label synthetic content
- Prohibited Practices and AI LiteracyArticles 4 and 5
- General-Purpose ModelsGPAI Obligations and Penalty System
Postponed due to the Omnibus
- High Risk · Appendix IIIEmployment, credit, essential services, biometrics…
- High-Risk Products · Annex IAI Integrated into Regulated Products
The postponement does not change what will need to be demonstrated: inventory, classification, human oversight, and evidence. It only changes the date.
