A new study published in the journal Cybersecurity describes a data-level prompt injection attack against graph neural networks, which are used in security-sensitive areas such as fraud detection, recommendation systems, and scientific research. Researchers at Henan University of Science and Technology, led by Mengying Yuan and Zhiyong Zhang, call the threat Graph Prompt Injection Attack, or GPIA. Unlike conventional backdoor attacks that require poisoning the pretraining pipeline or tampering with model weights, GPIA operates entirely at the downstream adaptation stage of graph prompt learning, a technique that keeps a large pretrained graph encoder frozen and tunes only a tiny set of prompt parameters. The attacker slips a small number of carefully crafted malicious graphs into the labeled dataset a downstream user collects to tune prompts, leaving the pretrained encoder pristine and the training algorithm untouched, while the learned prompts absorb an attacker-specified rule that lies dormant until the right structural pattern appears. Because the encoder is fixed, task-specific knowledge concentrates in the prompt parameters, creating a previously underexplored attack surface. GPIA unfolds in three stages, beginning with a compact prompt-conditioning subgraph optimized offline against the frozen encoder so graphs carrying it drift toward a chosen target representation while otherwise staying close to their clean semantics.