Why Safety-First AI Governance Risks Producing Unsafe Systems

Contemporary artificial intelligence governance is increasingly framed around safety-first principles emphasizing restriction, limitation, and centralized control. While such approaches aim to prevent misuse and harm, this paper argues that an overreliance on safety-centric constraint risks reproducing a broader institutional pattern: the suppression of human agency in the name of risk mitigation. Drawing on a structural analysis of safety theatre, the paper examines how AI governance frameworks may substitute procedural restriction for competence development, creating systems that appear controlled while becoming brittle under real-world complexity. We argue that safety in AI systems depends not only on limiting machine behavior, but on preserving and cultivating human judgment, interpretability, and intervention capacity. The paper concludes that agency is not an optional add-on to AI safety, but a foundational requirement for resilient oversight.