We investigate the detection prospects for gravitational lensing of three-dimensional maps from upcoming line intensity surveys, focusing in particular on the impact of gravitational nonlinearities on standard quadratic lensing estimators. Using perturbation theory, we show that these nonlinearities can provide a significant contaminant to lensing reconstruction, even for observations at reionization-era redshifts. However, we show how this contamination can be mitigated with the use of a "bias-hardened" estimator. Along the way, we present an estimator for reconstructing longwavelength density modes, in the spirit of the "tidal reconstruction" technique that has been proposed elsewhere, and discuss the dominant biases on this estimator. After applying bias-hardening, we find that a detection of the lensing potential power spectrum will still be challenging for the first phase of SKA-Low, CHIME, and HIRAX, with gravitational nonlinearities decreasing the signal to noise by a factor of a few compared to forecasts that ignore these effects. On the other hand, cross-correlations between lensing and galaxy clustering or cosmic shear from a large photometric survey look promising, provided that systematics can be sufficiently controlled. We reach similar conclusions for a single-dish survey inspired by CII measurements planned for CCAT-prime, suggesting that lensing is an interesting science target not just for 21cm surveys, but also for intensity maps of other lines.