Prey assessing risk may miss cues and fail to defend themselves, or respond unnecessarily to false alarms. Error rates can be ameliorated with more information, but sampling predator cues entails risk. Red-eyed treefrogs have arboreal eggs and aquatic tadpoles. The embryos use vibrations in snake attacks to cue behaviorally mediated premature hatching, and escape, but vibrations from benign sources rarely induce hatching. Missed cues and false alarms are costly; embryos that fail to hatch are eaten and hatching prematurely increases predation by aquatic predators. Embryos use vibration duration and spacing to inform their hatching decision. This information accrues with cycles of vibration, while risk accrues over time as snakes feed. We used vibration playback experiments to test if embryos adjust sampling of information based on its cost, and measured latency to initiate hatching in videotaped snake attacks. Embryos did not initiate hatching immediately in attacks or playbacks, and the delay varied with the rate at which information accrued. Embryos started hatching sooner in response to stimuli with shorter cycles but sampled fewer cycles (less information) of longer-cycle stimuli before hatching. This flexible sampling is consistent with embryos balancing a trade-off between the value and cost of information.