Safety or surveillance? The 'Safety exception' for neuroergonomic monitoring in the EU AI act

The European Commission's "human-centric" approach to Artificial Intelligence (AI) encourages the development of "Collaborative Intelligence" to enhance human capabilities, especially in safety-critical industries. This has spurred innovations in neuroergonomics and physiology-recording wearable sensors, designed to monitor operator states like fatigue, stress, and cognitive load to improve workplace safety. However, this same technology presents a profound risk of enabling unprecedented, invasive workplace surveillance. This paper identifies a critical, unresolved ambiguity at the nexus of this conflict: the EU's landmark AI Act. While the Act commendably bans "inferring emotions in workplaces" as an "unacceptable risk," it provides a vague exception for "medical or safety reasons." This "safety exception" is dangerously undefined and threatens to create a legal loophole for ethics-washing, where invasive surveillance is rebranded as a safety feature. This paper performs a conceptual analysis, arguing that this ambiguity cannot be resolved by the AI Act alone. We contend that the exception must be interpreted through a multi-layered legal framework integrating the stricter principles of the General Data Protection Regulation (GDPR) concerning health data, the emerging rules on algorithmic management from the Platform Directive, and the powerful co-determination rights of national labor laws, such as the German Works Constitution Act. We conclude by proposing a five-point operational test for regulators, developers, and Works Councils to distinguish between legitimate, safety-enhancing neuroergonomic systems and prohibited, disproportionate surveillance. This framework operationalizes the Platform Work Directive (2024/2831) provisions on algorithmic management and demonstrates how the AI Act's "humancentric" goals can be enforced through multi-layered legal accountability.

Keywords: EU AI Act, Neuroergonomics, Workplace Surveillance, Human-Centric AI, Safety-Critical Systems, GDPR, Algorithmic Management, Platform Work Directive, Codetermination, Emotion Recognition.