Earth's ocean-atmosphere system has undergone a dramatic but protracted increase in oxygen (O₂) abundance. This environmental transition ultimately paved the way for the rise of multicellular life and provides a blueprint for how a biosphere can transform a planetary surface. However, estimates of atmospheric oxygen levels for large intervals of Earth's history still vary by orders of magnitude-foremost for Earth's middle history. Historically, estimates of mid-Proterozoic (1.9-0.8 Ga) atmospheric oxygen levels are inferred based on the kinetics of reactions occurring in soils or in the oceans, rather than being directly tracked by atmospheric signatures. Rare oxygen isotope systematics-based on quantifying the rare oxygen isotope ¹⁷O in addition to the conventionally determined ¹⁶O and ¹⁸O-provide a means to track atmospheric isotopic signatures and thus potentially provide more direct estimates of atmospheric oxygen levels through time. Oxygen isotope signatures that deviate strongly from the expected mass-dependent relationship between ¹⁶O, ¹⁷O, and ¹⁸O develop during ozone formation, and these "mass-independent" signals can be transferred to the rock record during oxidation reactions in surface environments that involve atmospheric O₂. The magnitude of these signals is dependent upon 𝘱O₂, 𝘱CO₂, and the overall extent of biospheric productivity. Here, we use a stochastic approach to invert the mid-Proterozoic Δ¹⁷O record for a new estimate of atmospheric 𝘱O₂, leveraging explicit coupling of 𝘱O₂ and biospheric productivity in a biogeochemical Earth system model to refine the range of atmospheric 𝘱O₂ values that is consistent with a given observed Δ¹⁷O. Using this approach, we find new evidence that atmospheric oxygen levels were less than ∼1% of the present atmospheric level (PAL) for at least some intervals of the Proterozoic Eon.