Since August 2, 2026, the European AI Act has moved from text to enforcement. The transparency obligations set out in Article 50 are now fully applicable, and national authorities have effective investigative powers. For leaders deploying AI tools internally or in customer relations, the regulatory framework is no longer a distant horizon: it is an immediate operational constraint.
Sanctions of the AI Act: what companies risk concretely
The sanctions regime accompanying the enforcement of transparency obligations changes the nature of risk. Violations can be penalized up to 15 million euros or 3% of global annual revenue, with the higher amount being retained.
The European Commission and national authorities are no longer just issuing rules. They have inspection, investigation, and deployment restriction powers that are now being used effectively.
For companies using generative AIs (text, image, video), two practical obligations arise. First, to make content identifiable as AI-generated. Second, to clearly inform users that they are interacting with an artificial intelligence, not a human. These requirements apply to both public-facing chatbots and those deployed internally.
Several decision-makers follow the news on Le Blog des Décideurs to anticipate these regulatory developments and adapt their processes before an audit.
Data governance and compliance: the priority project for leaders
The AI Act is just one brick among others. The European directive on pay transparency, with deadlines approaching, requires companies to structure and validate their HR data. Cross-referencing these two regulatory requirements reveals a common problem: most internal information systems are not ready for a compliance audit.
Salary data, application sorting algorithms, internal scoring tools: every technological layer used in talent management must be documentable and explainable. Field feedback varies on this point, as some SMEs have already begun this mapping work while others are discovering the extent of the task.
What compliance requires in practice
- A comprehensive inventory of AI systems used in HR processes (recruitment, evaluation, internal mobility), with technical documentation accessible to regulatory authorities.
- Reporting mechanisms integrated into user interfaces so that every employee or candidate knows when they are interacting with an algorithm.
- A transversal governance involving legal, IT, and HR departments, because compliance cannot rest on a single department.
Companies treating this issue as an isolated IT project expose themselves to blind spots. Compliance with the AI Act affects both external communication and internal skills management processes.
Agentic AI in business: between operational promise and practical limits
Beyond regulation, the technology itself is evolving. Agentic AI, capable of autonomously executing complex chained tasks, is beginning to deploy in support functions. Expense management, pre-qualification of applications, activity report synthesis: use cases are multiplying.
The available data does not yet allow for conclusions on real productivity gains at scale. However, companies managing these deployments are already measuring a gap between the promises of vendors and operational reality. The autonomy of these agents remains conditioned by the quality of input data and the clarity of the business rules governing them.
Skills and training in the face of automation
The arrival of agentic AI redefines the skills expected from employees. Middle managers find themselves on the front lines: their role is evolving from supervising tasks to supervising software agents.
Training can no longer be treated as an annual catalog. It becomes a continuous investment, focused on the ability to configure, audit, and correct decisions made by autonomous systems. HR teams that continue to think in terms of training hours are missing this shift.
Mental health and employee engagement: a management indicator, not a program
Mental health at work has moved from discourse to measurement. Several organizations are now integrating perceived workload and quality of life indicators into their management dashboards, alongside revenue or turnover rates.
This change in status has a direct consequence on team management. When mental health becomes an indicator monitored by the management committee, managers receive a clear mandate to arbitrate between workload and production goals. Without this mandate, well-being policies remain declarative.
- Integrate a perceived health indicator into monthly reporting, linked to absenteeism and turnover data, to objectify alerts.
- Train managers to detect weak signals (gradual disengagement, isolation in a hybrid context) rather than just crisis management.
- Distinguish structural prevention actions (workload, decision-making autonomy) from individual support measures, which only address the consequences.
Companies that approach mental health as a distinct HR program, disconnected from operational management, generally see a low impact on actual employee engagement.
The second half of 2026 places decision-makers in the face of simultaneous trade-offs: regulatory compliance on AI, restructuring of internal skills, and integration of health at work into strategic management. These three projects are not independent. The quality of data governance conditions both legal compliance and the relevance of decision-support tools, whether they concern talent or performance.
