Metraipolis › archive › dc_fd15215f382f
Rethinking Assessments of Prompt Injection Attacks (2026) · OpenAlex W7166030222
- Source
- OpenAlex
- Original
- https://doi.org/10.60882/cispa.32802851.v1
- License (as archived)
- CC-BY-4.0
- Archived
- 2026-10-07T10:44:46.352Z
- SHA-256 of the archived text
31a0ef817092e9a1b9a0bbdc3842c5c0b1f0e8b83d20578f9ad17358ce6b67c1- Stable URL of this page
https://metraipolis.com/documento/dc_fd15215f382f
How to cite
Rethinking Assessments of Prompt Injection Attacks (2026) · OpenAlex W7166030222 — OpenAlex, https://doi.org/10.60882/cispa.32802851.v1 (CC-BY-4.0); archived copy: https://metraipolis.com/documento/dc_fd15215f382f (Metraipolis archive dc_fd15215f382f, sha256 31a0ef817092e9a1b9a0bbdc3842c5c0b1f0e8b83d20578f9ad17358ce6b67c1)
BibTeX
@misc{dcfd15215f382f,
title = {Rethinking Assessments of Prompt Injection Attacks (2026) · OpenAlex W7166030222},
howpublished = {OpenAlex},
url = {https://doi.org/10.60882/cispa.32802851.v1},
note = {License CC-BY-4.0; archived copy https://metraipolis.com/documento/dc_fd15215f382f; sha256 31a0ef817092e9a1b9a0bbdc3842c5c0b1f0e8b83d20578f9ad17358ce6b67c1},
year = {2026}
}Archived text
Reproduced under CC-BY-4.0. Attribution: OpenAlex, https://doi.org/10.60882/cispa.32802851.v1. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Opera scientifica in accesso aperto: scheda dai metadati OpenAlex (conference-paper). Titolo: Rethinking Assessments of Prompt Injection Attacks Autori: Chi Cui, Yixin Wu, Michael Backes, Yang Zhang Anno: 2026 · Sede: CISPA Helmholtz Center DOI: https://doi.org/10.60882/cispa.32802851.v1 Accesso aperto: green · testo: https://doi.org/10.60882/cispa.32802851.v1 · licenza del testo: cc-by Citazioni ricevute (OpenAlex): 0 Temi (OpenAlex): Adversarial Robustness in Machine Learning; Web Application Security Vulnerabilities; Security and Verification in Computing Abstract (testo degli autori, licenza cc-by): Prompt injection attacks are recognized as one of the primary risks faced by LLM-integrated applications in recent years. However, common evaluation frameworks remain insufficient, lacking comprehensiveness and real-world relevance. To bridge this gap, we revisit the common evaluation framework and conduct an extensive evaluation across eight different evaluation settings, including 37 real-world applications, 185 injected tasks, 21 attack instructions, and a total of 143,745 queries. The evaluation highlights several findings. For example, real-world applications are more vulnerable to prompt injection attacks compared to those used in research settings. While complex attack instructions are more sophisticated, they are less effective than simple attack instructions. To uncover the root causes of these phenomena, we further investigate the model’s internal representations during attacks, offering profound insights into the underlying dynamics of these attacks. Additionally, we conduct an assessment of both prompt-level and model-level defense mechanisms and highlight their limitations in real-world applications. By exploring more diverse scenarios across different dimensions, our framework provides a solid foundation for assessing vulnerabilities in LLM-integrated applications and evaluating the efficacy of defensive strategies. Fonte: https://openalex.org/W7166030222 — metadati OpenAlex (CC0 1.0), consultati il 2026-10-06 via api.openalex.org. Interrogazione: «prompt injection attacks».