We prove upper bounds on the one-arm exponent $η_1$ for a class of dependent percolation models which generalise Bernoulli percolation; while our main interest is level set percolation of Gaussian fields, the arguments apply to other models in the Bernoulli percolation universality class, including Poisson-Voronoi and Poisson-Boolean percolation. More precisely, in dimension $d=2$ we prove that $η_1 \le 1/3$ for continuous Gaussian fields with rapid correlation decay (e.g. the Bargmann-Fock field), and in $d \ge 3$ we prove $η_1 \le d/3$ for finite-range fields, both discrete and continuous, and $η_1 \le d-2$ for fields with rapid correlation decay. Although these results are classical for Bernoulli percolation (indeed they are best-known in general), existing proofs do not extend to dependent percolation models, and we develop a new approach based on exploration and relative entropy arguments. The proof also makes use of a new Russo-type inequality for Gaussian fields, which we apply to prove the sharpness of the phase transition and the mean-field bound for finite-range fields.