On May 19, 2026, the European Commission published its draft guidelines on the classification of high-risk artificial intelligence (“AI”) systems under the EU Artificial Intelligence Act. Alongside the announcement, a public consultation was launched, open until June 23, 2026. While the document is not yet legally binding—with the Court of Justice of the European Union holding the authority for final interpretation—the draft provides crucial insight into the regulator’s intentions and, more importantly, amends compliance deadlines.
What’s in the Draft Guidelines?
One of the most critical aspects of the EU AI Act, which entered into force on August 1, 2024, is the definition of “high-risk” systems. The new draft guidelines, issued under Article 6(5) of the Act, aim to clarify this very issue. The document is structured in three main parts:
Do you have a question about AI security? You can reach us here:
- Section 2 addresses classification under Article 6(1) and Annex I of the EU AI Act.
- Section 3 covers systems falling under Article 6(2) and Annex III. This includes certain stand-alone AI systems used in sensitive areas such as biometrics, education, employment, access to essential services, and law enforcement.
The draft provides non-exhaustive examples of AI systems that may or may not be classified as high-risk. This approach offers flexibility but also introduces a degree of uncertainty for companies until the final version is adopted following the consultation.
Extended Deadlines: Can Companies Breathe a Sigh of Relief?
Guidance on high-risk classification was originally expected by February 2, 2026. With the current publication, the “Digital Omnibus on AI” has also revised the implementation schedule, which is perhaps the most significant news for the market. The updated timetable is as follows:
- Requirements for stand-alone high-risk AI systems are now due to apply from December 2, 2027.
- For high-risk AI systems embedded in products, compliance obligations will apply from August 2, 2028.
From an AIQ standpoint, this extension is critical, but it is an opportunity for strategic preparation, not for inaction. Classifying an AI system as high-risk is a complex legal and technical task that requires in-depth analysis. Companies should use this extra time to conduct internal audits, identify potentially affected systems, and build the necessary internal compliance frameworks. The current draft acts as a compass, helping to fine-tune product development and risk management strategies.
The AI Act and the OWASP LLM Top 10: Security in Practice
In a corporate context, this means that legal compliance and cybersecurity are deeply intertwined. While the AI Act defines what systems are high-risk, frameworks like the OWASP LLM Top 10 provide answers as to how these systems can be vulnerable.
From AIQ’s perspective, a high-risk classification dramatically increases the gravity of technical vulnerabilities. Let’s consider a few examples:
- Employment (Annex III): For a high-risk AI system that pre-screens resumes, a successful Prompt Injection (LLM01) attack is not just a technical glitch but a severe discrimination and legal risk, violating both the AI Act and GDPR provisions.
- Biometrics (Annex III): If Sensitive Information Disclosure (LLM06) occurs in such a system, it is not merely a data breach but a violation of fundamental rights as stipulated by the AI Act.
- Essential Services (Annex III): In a credit scoring system, a flawed decision caused by Insecure Output Handling (LLM02) can have direct and severe consequences for individuals, establishing grounds for liability under the Act.
Therefore, a high-risk classification means in practice that LLM red teaming and vulnerability assessments are no longer just “best practices” but essential tools for legal compliance.
Audit and Compliance: What to Do Now?
The opportunity provided by the extended deadlines must be seized. From an AIQ standpoint, companies should consider the following steps:
- Internal Assessment: Immediately begin mapping existing and in-development AI systems against the criteria in the draft guidelines. Which systems might fall under Annex III? Which ones perform activities that carry high risk?
- Risk Analysis and Audit: For potentially high-risk systems, conduct a thorough technical and compliance risk assessment. This includes security audits, LLM red teaming, and bias testing to uncover hidden vulnerabilities.
- Build Documentation and Processes: The AI Act imposes strict documentation requirements. This period is perfect for establishing the internal processes that ensure ongoing compliance and accountability, in line with GDPR principles.
- Participate in the Consultation: Relevant industry stakeholders should take the opportunity to provide feedback on the draft before the June 23, 2026 deadline, to help ensure the final regulation is as practical as possible.
The extended preparation time is a valuable resource, but the task is complex. A proactive, security-first approach will not only help avoid fines but also build trust with users and customers in an increasingly AI-driven world.