The environmental dissemination of antibiotic resistance genes (ARGs) is a cornerstone of the One Health crisis, linking human, animal, and environmental health. Engineered ecosystems, such as constructed wetlands (CWs), are critical for water purification, but the design of their core reactive media has focused on pollutant removal, overlooking potential ecological risks. This study reveals a fundamental performance-risk trade-off by comparing four biogeochemically distinct substrates (river sand, zeolite, biochar, pyrite) in CWs treating multi-antibiotic wastewater. While biochar-amended CWs excelled at removing COD (95.2 ± 2.1%) and parent antibiotics (>90%), they simultaneously evolved into hotspots for antibiotic resistance, elevating the horizontal transfer potential (intI1 gene) by approximately 340-fold and enriching for multiple ARGs compared with control. Furthermore, biochar steered antibiotic degradation towards transformation products with a predicted potential for significantly higher ecotoxicity towards Daphnia. In contrast, pyrite-amended CWs demonstrated a paradigm of safe efficiency: despite exhibiting comparable high removal rates for organic matter and antibiotics, they simultaneously suppressed ARG proliferation via a biogeochemical stress mechanism driven by iron-sulfur cycling and elevated reactive oxygen species (ROS) production and minimized nitrous oxide (N2O) emissions by 78% relative to the control. Machine learning and structural equation modeling identified distinct microbial functional guilds as the causal drivers of these divergent risks. Our findings demonstrate that prioritizing removal efficiency alone is a flawed strategy. Pyrite emerges as a superior functional substrate that optimally balances high performance with low ecological risk, providing a sustainable, One Health-aligned engineering solution for the design of next-generation, environmentally safe treatment bioreactors.