Global ocean models that admit mesoscale turbulence spontaneously generate a substantial interannual-to-multidecadal chaotic intrinsic variability in the absence of atmospheric forcing variability at these timescales. is phenomenon is substantially weaker in non-turbulent ocean models but provides a marked stochastic avor to the low-frequency variability in eddying ocean models, which are being cou- pled to the atmosphere for next-generation climate projections. In order to disentangle the atmospherically forced and intrinsic ocean variabilities, the OCCIPUT (OceaniC Chaos – ImPacts, strUcture, predicTability) project performed a long (1960–2015), large ensemble (50 members) of global ocean/sea ice 1/4° simulations driven by the same atmospheric reanalysis, but with perturbed initial conditions. Subsequent ensem- ble statistics show that the ocean variability can be seen as a broadband “noise,” with characteristic scales reaching multiple decades and basin sizes, locally modulated by the atmospheric variability. In several mid-latitude regions, chaotic processes have more impact than atmospheric variability on both the low-frequency variability and the long-term trends of regional ocean heat content. Consequently, certain climate- relevant oceanic signals cannot be unambiguously attributed to atmospheric variabil- ity, raising new issues for the detection, attribution, and interpretation of oceanic heat variability and trends in the presence of mesoscale turbulence.