This document proposes temporal defenses for robust recommendations to detect sybil attacks on recommender systems. It suggests (1) distrusting newcomers to force sybils to appear more than once, (2) examining sybil group dynamics like number of sybils and ratings per sybil over time, (3) monitoring at the system, user, and item level for anomalous changes, and (4) flagging changes that exceed thresholds to force sybils to attack more intelligently. The approach is evaluated using simulations and injecting real data with attacks.