The amended safety-component test provides greater clarity for AI systems associated with regulated products
Under the EU AI Act, AI systems classified as ‘high-risk’ face significant obligations - from technical documentation and quality management systems to third-party conformity assessments. For many manufacturers, particularly those in regulated sectors such as medical devices, a key question has been: whether the AI component fulfils a safety function?
The AI Omnibus has now provided much-needed clarity by narrowing the definition of what constitutes a ‘safety component’ - one of the key triggers for high-risk classification.
What changed and why does it matter?
Under the original AI Act, an AI system is classified as high-risk where it is itself a regulated product or where it serves as a safety component of such a product and is subject to third-party conformity assessment. The AI Omnibus seeks to address this concern by clarifying that Annex I classification should remain focused on AI functions that are genuinely relevant to product safety.
It introduces new language in Article 6(1) to clarify what is not considered a safety function:
“Systems that are solely used for non-safety related aspects of user assistance, performance optimization, service efficiency, automation or convenience or quality control shall not qualify as safety components.” - Article 6(1a), EU AI Act as amended by the AI Omnibus
In practice, this means that AI functions designed primarily to improve usability, workflow efficiency, or user experience - rather than to perform a safety-critical role - may no longer automatically fall within the high-risk regime, even when embedded within a regulated product such as a medical device.
What does this mean for medical device manufacturers?
This is likely to be one of the most significant practical consequences of the AI Omnibus for manufacturers in regulated sectors.
Previously, there was a concern that any AI-enabled feature within a medical device could trigger the full suite of high-risk AI obligations. The clarification shifts the focus to the intended purpose and function of the AI system. Manufacturers will now need to ask: does this AI function perform a safety-related role? Could its failure affect the health and safety of people or property?
Under AI omnibus, AI systems used solely for specified non-safety-related functions do not qualify as safety components. However, an AI system whose failure or malfunction would endanger health and safety qualifies as a safety component. The other conditions, including the applicable third-party conformity assessment requirement, must still be considered.
Reducing duplication with existing sectoral legislation
The AI Omnibus also takes steps to reduce the compliance burden on manufacturers who are already subject to robust regulatory requirements through existing EU legislation, such as the Medical Devices Regulation (MDR) or the In Vitro Diagnostic Medical Devices Regulation (IVDR).
A key new provision allows certain AI Act requirements to be limited where equivalent requirements already exist under sectoral legislation and provide a comparable level of protection. The European Commission is required to adopt delegated acts by 2 August 2027 to define exactly which requirements this applies to and how the mechanism will work in practice.
Guidance on the practical implementation of AI Act requirements for these products is expected to be published by 1 August 2027.
When do the high-risk AI requirements apply to Annex I products?
The application date for high-risk AI systems covered by Annex I (which includes products regulated under Union harmonization legislation such as medical devices) has been extended to 2 August 2028, giving manufacturers, notified bodies, and regulators additional time to prepare.
High-risk biometric AI conformity assessment
Providers of high-risk biometric AI systems must follow the applicable AI Act conformity assessment procedure before placing the system on the market or putting it into service.
BSI is currently developing its conformity-assessment capability for high-risk biometric AI systems through its pilot programme as part of its preparation for AI Act Notified Body designation. Where Notified Body involvement is required or selected under the applicable AI Act conformity assessment route, a designated Notified Body may assess the QMS and technical documentation in accordance with the applicable procedure.
The service applies to Annex III, point 1 biometric AI systems under the EU AI Act, including:
- Remote biometric identification systems
- Biometric categorization systems based on sensitive or protected attributes
- Emotion recognition systems
Explore High-Risk Biometric Conformity Assessments.
Steps you can take for the future of AI Governance
The question now is a practical one. How can you demonstrate robust AI governance and provide evidence about how your AI performs?
Here are steps you can take to answer it.
AI Performance
Independent, evidence-based testing of your AI models and datasets. We verify accuracy, bias, and robustness, then you can display the BSI Mark of Trust. It's a clear signal to customers, investors, and regulators that your claims stand up.
ISO/IEC 42001 Pre-Certification Assessment
An informal, preliminary assessment of your organization’s AI Management System, to determine scope, readiness, and capability for an ISO/IEC 42001 certification audit.
ISO/IEC 42001 Certification
Certification of your AI Management System builds stakeholder trust, brings structure to how you develop and deploy AI, and demonstrates accountability through clear responsibilities for AI decisions and oversight.
Explore ISO/IEC 42001 certification.
To understand every key change in full detail, download the AI Omnibus: An Overview Recap. Written by our AI Regulatory Leads, it is your essential guide to what has changed and what to do next.
For an in-depth understanding of the EU AI Act, download the AI Act – What AI Providers and Deployers Need to Know whitepaper. Written by our AI Regulatory Leads, this white paper provides expert guidance on the requirements and implications of the EU AI Act.