Abstract
Temporal Graph Neural Networks (TGNNs) are increasingly used in high-stakesdomains, such as financial forecasting, recommendation systems, and frauddetection. However, their susceptibility to poisoning attacks poses a criticalsecurity risk. We introduce LoReTTA (Low Resource Two-phase Temporal Attack), anovel adversarial framework on Continuous-Time Dynamic Graphs, which degradesTGNN performance by an average of 29.47% across 4 widely benchmark datasets and4 State-of-the-Art (SotA) models. LoReTTA operates through a two-stageapproach: (1) sparsify the graph by removing high-impact edges using any of the16 tested temporal importance metrics, (2) strategically replace removed edgeswith adversarial negatives via LoReTTA's novel degree-preserving negativesampling algorithm. Our plug-and-play design eliminates the need for expensivesurrogate models while adhering to realistic unnoticeability constraints.LoReTTA degrades performance by upto 42.0% on MOOC, 31.5% on Wikipedia, 28.8%on UCI, and 15.6% on Enron. LoReTTA outperforms 11 attack baselines, remainsundetectable to 4 leading anomaly detection systems, and is robust to 4 SotAadversarial defense training methods, establishing its effectiveness,unnoticeability, and robustness.