Common integrated assessment models produce the counterintuitive result that higher risk aversion does not lead to stronger near-term abatement. This paper re-examines this result with a DICE model that is fully coupled with a thermohaline circulation model. It also features Epstein-Zin utility and uncertainty about climate sensitivity that resolves after some time. The simulations show that aversion to this tipping point risk has little effect. For climate sensitivity of realistic magnitude, a collapse of the circulation occurs in the distant future, which allows acting after learning. Furthermore, the anticipated damage costs are not sufficiently great to justify precautionary measures.