AIA Article 15. Accuracy, robustness and cybersecurity

High-risk AI systems should perform consistently throughout their lifecycle and meet an appropriate level of accuracy, robustness and cybersecurity, in light of their intended purpose and in accordance with the generally acknowledged state of the art. The Commission and relevant organisations and stakeholders are encouraged to take due consideration of the mitigation of risks and the negative impacts of the AI system. The expected level of performance metrics should be declared in the accompanying instructions of use. Providers are urged to communicate that information to deployers in a clear and easily understandable way, free of misunderstandings or misleading statements. Union law on legal metrology, including Directives 2014/31/EU and 2014/32/EU of the European Parliament and of the Council, aims to ensure the accuracy of measurements and to help the transparency and fairness of commercial transactions. In that context, in cooperation with relevant stakeholders and organisation, such as metrology and benchmarking authorities, the Commission should encourage, as appropriate, the development of benchmarks and measurement methodologies for AI systems. In doing so, the Commission should take note and collaborate with international partners working on metrology and relevant measurement indicators relating to AI.

Technical robustness is a key requirement for high-risk AI systems. They should be resilient in relation to harmful or otherwise undesirable behaviour that may result from limitations within the systems or the environment in which the systems operate (e.g. errors, faults, inconsistencies, unexpected situations). Therefore, technical and organisational measures should be taken to ensure robustness of high-risk AI systems, for example by designing and developing appropriate technical solutions to prevent or minimise harmful or otherwise undesirable behaviour. Those technical solution may include for instance mechanisms enabling the system to safely interrupt its operation (fail-safe plans) in the presence of certain anomalies or when operation takes place outside certain predetermined boundaries. Failure to protect against these risks could lead to safety impacts or negatively affect the fundamental rights, for example due to erroneous decisions or wrong or biased outputs generated by the AI system.

Cybersecurity plays a crucial role in ensuring that AI systems are resilient against attempts to alter their use, behaviour, performance or compromise their security properties by malicious third parties exploiting the system’s vulnerabilities. Cyberattacks against AI systems can leverage AI specific assets, such as training data sets (e.g. data poisoning) or trained models (e.g. adversarial attacks or membership inference), or exploit vulnerabilities in the AI system’s digital assets or the underlying ICT infrastructure. To ensure a level of cybersecurity appropriate to the risks, suitable measures, such as security controls, should therefore be taken by the providers of high-risk AI systems, also taking into account as appropriate the underlying ICT infrastructure.

Without prejudice to the requirements related to robustness and accuracy set out in this Regulation, high-risk AI systems which fall within the scope of a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements, in accordance with that regulation may demonstrate compliance with the cybersecurity requirements of this Regulation by fulfilling the essential cybersecurity requirements set out in that regulation. When high-risk AI systems fulfil the essential requirements of a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements, they should be deemed compliant with the cybersecurity requirements set out in this Regulation in so far as the achievement of those requirements is demonstrated in the EU declaration of conformity or parts thereof issued under that regulation. To that end, the assessment of the cybersecurity risks, associated to a product with digital elements classified as high-risk AI system according to this Regulation, carried out under a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements, should consider risks to the cyber resilience of an AI system as regards attempts by unauthorised third parties to alter its use, behaviour or performance, including AI specific vulnerabilities such as data poisoning or adversarial attacks, as well as, as relevant, risks to fundamental rights as required by this Regulation.

The conformity assessment procedure provided by this Regulation should apply in relation to the essential cybersecurity requirements of a product with digital elements covered by a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements and classified as a high-risk AI system under this Regulation. However, this rule should not result in reducing the necessary level of assurance for critical products with digital elements covered by a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements. Therefore, by way of derogation from this rule, high-risk AI systems that fall within the scope of this Regulation and are also qualified as important and critical products with digital elements pursuant to a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements and to which the conformity assessment procedure based on internal control set out in an annex to this Regulation applies, are subject to the conformity assessment provisions of a regulation of the European Parliament and of the Council on horizontal cybersecurity requirements for products with digital elements insofar as the essential cybersecurity requirements of that regulation are concerned. In this case, for all the other aspects covered by this Regulation the respective provisions on conformity assessment based on internal control set out in an annex to this Regulation should apply. Building on the knowledge and expertise of ENISA on the cybersecurity policy and tasks assigned to ENISA under the Regulation (EU) 2019/881 of the European Parliament and of the Council , the Commission should cooperate with ENISA on issues related to cybersecurity of AI systems.