I will review a puzzle appearing in the AdS/CFT correspondence where products of CFT partition functions don’t factorize when computed from the bulk due to the presence of Euclidean wormholes. It is currently an open question on whether these wormholes should contribute to the path integral for definite microscopic CFTs where no disorder average is taken. In an attempt to explain the appearance of these wormholes, I will propose an ansatz for OPE coefficients of chaotic CFTs that generalizes the Eigenstate Thermalization Hypothesis and treats heavy states as random variables. Applying this ansatz to compute the square of a genus-2 partition function will reproduce the contribution of the genus-2 wormhole described by Maoz and Maldacena many years ago. Finally, I will argue that gravitational computations using the low-energy effective action can only ever capture the random nature of the OPE coefficients.