LES ACTIVITÉS NUCLÉAIRES : RAYONNEMENTS IONISANTS ET RISQUES POUR LA SANTÉ ET L’ENVIRONNEMENT the mandatory controls have actually been carried out and, on the other, enables the licensees to be questioned about the steps taken to remedy any nonconformities. In 2023, a large number of RNAs had not identified the need to have checks required by the Public Health Code performed by an Approved Organisation for Radiation Protection (AORP). During the course of 2023, ASN issued communications on the regulatory requirements, notably via the Radiation Protection Expert-Officer (RPE-O) networks and conducted an information campaign (by post) targeting the RNAs concerned. These actions have been effective. The number of checks carried out under the Public Health Code doubled in all sectors in 2024. The breakdown per field is given in Table 2. The reports of the verifications performed in each facility by the AORP verifications are at the disposal of and examined by ASNR personnel on the occasion of: ∙licence renewals or modifications requiring ASNR authorisation; ∙inspections. ASNR also approves laboratories to conduct analyses requiring a high level of measurement quality if the results are to be usable. It thus approves laboratories to monitor radioactivity in the environment (see point 5.3). In addition, ASNR approves following the advice of the standing subcommittee in charge of the transport of hazardous goods within the High Council for the Prevention of Technological Risks (CSPRT): ∙the training organisations for drivers of vehicles carrying radioactive materials; two organisations have been approved; ∙the organisations responsible for certifying the conformity of packagings designed to contain 0.1 kilogramme (kg) or more of uranium hexafluoride (UF6); ∙the organisations responsible for type approval of tank containers and swap tanks intended for the carriage of class 7 dangerous goods; ∙the organisations responsible for the initial and periodic inspections of tanks intended for the carriage of class 7 dangerous goods. Two organisations are approved for the qualification of tankcontainers and for certification of the conformity of UF6 packagings. As at 31 December 2025, the following are approved or accredited by ASNR: ∙9 organisations responsible for radiation protection verifications. One approval renewal was delivered in 2025; ∙67 organisations responsible for measuring radon activity concentration in buildings (level 1), 14 of which are also approved to identify sources and routes for entry and transfer of radon in buildings (level 2). In 2025, 37 new approvals or approval renewals were issued, 32 of which were level 1 and 5 were level 2; ∙4 organisations qualified for NPE inspections as part of the new NPE conformity assessment; ∙2 organisations qualified for NPE inspections as part of in-service monitoring; ∙3 organisations qualified for PE and simple pressure vessels within the scope of BNIs (in-service monitoring); ∙18 inspection departments qualified for in-service monitoring of NPE and simple pressure vessels within the scope of NPPs; ∙69 laboratories approved for environmental radioactivity measurements covering 988 approvals currently valid as at 1 January 2026, of which 285 are approvals or approval renewals delivered or maintained during the course of 2025. Since 2020, the regulations have gradually restricted the scope of intervention of the AORPs by delegating the verification duties set out in the Labour Code to verification organisations accredited Medical Research / Teaching Industry (Including Veterinary) Total 213 84 75 372 * Given the changes to the regulations, the number of checks performed is now expressed in the number of facilities checked (rather than the number of sources). TABLE 2 Radiation protection verifications performed in 2025 by the organisations approved for radiation protection verifications(*) HIGHLIGHT No. 1 Strategic issues linked to the use of artificial intelligence in nuclear facilities Artificial Intelligence (AI) encompasses a range of techniques, some of which have long been used in the nuclear sector, in particular so-called “symbolic” AI. These techniques, which are based on explicit rules, enable the reasoning carried out by the systems to be traced and the results obtained to be made explicit. Based on logical rules and reasoning, these applications have been used as decision aids for diagnosing the operation of nuclear facilities, for example to identify losses of electrical sources, analyse reactor coolant pump faults or monitor the vibration of turbine-generator sets. In recent years, other approaches to AI, particularly those based on data-driven learning, have seen rapid growth in industrial sectors, including nuclear power. Although their use for security purposes is still limited, their deployment is set to increase. Today, these technologies are used, for example, to support maintenance, monitoring and analysis of plant operation. The tools currently being developed complement existing organisations and are under human control. They are usually based on hybrid approaches combining physical models, formalised domain expertise and data analysis. At this stage, these developments represent an opportunity to improve the early detection of faults and maintain the compliance of facilities with their safety standards. However, the eventual move towards more autonomous systems, replacing historical approaches, would constitute a safety issue. As with the introduction of industrial IT in nuclear safety in the past, the growing use of AI raises fundamental issues: robustness and justification of results, control of system complexity, understanding of system behaviour in degraded or unforeseen situations, as well as supervision, oversight, validation and qualification of models. These challenges are particularly acute for learning-based approaches, which are heavily dependent on data, its representativeness for the system’s field of application and its context of use. The quality, governance and representativeness of the data thus appear to be particularly key factors, especially for rare, poorly documented events. What’s more, the integration of AI in sensitive industrial environments raises the stakes in terms of cybersecurity and protecting the integrity of digital systems. Against this backdrop, ASNR has embarked on the “Augmented Expertise” programme, aimed both at strengthening its expertise capabilities through the controlled use of AI and to plan ahead for evaluation of tools developed in the future by industry using new AI techniques. In this context, ASNR has already developed innovative expert assessment systems designed as expert assessment aids, under expert oversight. These systems are specifically intended to make it easier to identify configurations with safety implications, to which the expert must pay particular attention. At the same time, at national level, ASNR is engaging with companies on topics ranging from benchmarking AI methods and techniques to the drafting, with EDF, of a white paper on the use of AI in safety, scheduled for publication in 2026. At international level, discussions are also held with other safety authorities, in particular as part of dedicated working groups or bilateral cooperation (NRC in the United States and GRS in Germany) and multilateral cooperation (IAEA, NEA). The aim of this approach is to plan ahead for technological developments, prepare the regulatory framework and ensure that the use of AI is compatible with the stringent requirements of nuclear safety. ASNR Report on the state of nuclear safety and radiation protection in France in 2025 145 01 02 03 05 09 06 10 07 11 13 04 08 12 A / Z
RkJQdWJsaXNoZXIy NjQ0NzU=