Gas-rich hot springs throughout the Peruvian Andes contain a surprising contribution of mantle and crustal volatiles (CO2 and N2) despite being located along a volcanic gap associated with modern flat-slab subduction. Similar mantle and crustal volatile degassing is observed in springs from the backarc region of the northern Altiplano Plateau which experienced flat-slab subduction in the Oligocene. We constrain the sources of deeply-derived volatiles using C, N, O and H stable isotope analyses alongside previously reported helium isotope (3He/4He) and gas abundance data. δ18O and δ2H values of spring waters are consistent with regional meteoric water, with deviations towards higher values due to fluid mixing and high-temperature fluid-rock interaction. δ15N values of N2 gas range from −0.5 to +4.4‰ (vs AIR), indicating contributions of mantle and crustal (sedimentary) nitrogen. δ13C values of dissolved inorganic carbon (DIC) and CO2 gas from bubbling springs are used to model initial δ13C values before degassing, with resulting values ranging from −13.6 to −0.3‰ (vs VPDB) and an average value of −6.9‰. DIC concentrations range from 1 to 57 mmol/kg, of which carbonate dissolution accounts for only a small fraction (average of 15%). The remaining carbon (1 to 44 mmol/kg) is derived from a mixture of deep CO2 sources with endmember δ13C values of −7‰ and −14‰ which, in light of relationships with 3He/4He ratios, are interpreted to be the subcontinental lithospheric mantle (SCLM) and metamorphic crustal carbon, respectively. These results are consistent with previous interpretations that slab-to-lithosphere fluid transfer is mobilizing volatiles from the SCLM to the continental crust above the flat slab and confirms that mantle CO2 is migrating with mantle helium. Metamorphic CO2 and N2 are mobilized from the continental crust and entrained by mantle fluids on their way to the surface. In the backarc region, similar geochemical patterns suggest that volatiles are released by dehydration and/or partial melting of previously hydrated lithosphere. We estimate a CO2 flux of ~3 × 108 mol yr−1 (34 t d−1) from 45 springs with available discharge values. If scaled to all reported thermal springs in Peru, then the total CO2 emissions from Peruvian thermal springs may approach 2 × 109 mol yr−1 (203 t d−1), or ~ 0.2% of global emissions from subaerial volcanism. Regional CO2 emissions may be 10 to 100 times greater when considering CO2 diffusely lost along fault zones or temporarily dissolved near-surface aquifers. These results demonstrate that flat-slab subduction leads to an efficient and unexpected transfer of mantle and crustal volatiles to Earth's surface, and more generally, that deep volatile fluxes in subduction zones are not limited to active volcanism.