Abstract
<jats:p>The global expansion of genetically modified plants supports modern food and ecological security, yet whether the fitness cost (widely assumed to limit their spread after escape) actually operates in the wild has remained unclear, with empirical evidence remaining mixed. Here, a global meta-analysis of 518 experiments across 48 species reveals that this safety brake is real in the laboratory (fitness cost: ~13%) but is largely attenuated in the world's most intensively cultivated croplands. A phylogenetically robust, hump-shaped relationship between mean annual temperature and fitness effect is defined, with fitness benefits emerging specifically within 9.9 to 22.8 celsius centigrade. This buffer window overlaps the productive croplands, where the high-risk proportion (HRP; the area-weighted posterior probability of non‑negative fitness effect) consistently exceeds that in natural vegetation and expands with global warming (cropland: +18.4% versus natural vegetation: +8.3% under SSP5-8.5). Notably, constitutive expression or dual-resistance (resist to both abiotic and biotic stressors) transgene can effectively compress the HRP below 30%, providing a quantitative baseline for management. These findings establish a framework for climate-smart deployment that balances agricultural productivity with ecological safety.</jats:p>