Newest papers

Adversarial Benchmarks and BenchDoor Attacks: Exploiting Benchmark Incentives for Backdoor Attacks

Yixin Wu, Rachel S. Hong, Jamie Morgenstern, Tadayoshi Kohno · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.27533 · cited by 0 works (OpenAlex) · open-access PDF

French

Nous présentons BenchDoor, une famille d’attaques d’intoxication de données qui exploite cette incitation. Au-delà de la préservation de la performance bénigne, BenchDoor exige des gains de référence qui encouragent l’adoption de données qui implantent également une porte arrière. L’adversaire manipule soit la formation d’un référentiel, soit les données de test, crée un référentiel adversaire, soit fournit un ensemble de données d’entraînement indépendant tout en laissant le référentiel inchangé. Nous formalisons cinq scénarios d’attaque, y compris la formation sur les exemples de tests réécrits et l’application de la décontamination basée sur la similitude. À travers un Q&A multi-choix, une classification de texte et un Q

Machine translation (Meta M2M100); the English original follows.

Abstract

Machine learning benchmarks act as guideposts for model development and spur the creation of training datasets intended to improve benchmark scores. We introduce BenchDoor, a family of data poisoning attacks that exploits this incentive. Beyond preserving benign performance, BenchDoor requires benchmark gains that encourage adoption of data that also implants a backdoor. The adversary either manipulates a benchmark's training or test data, creating an adversarial benchmark, or supplies an independent training dataset while leaving the benchmark unchanged. We formalize five attack scenarios, including training on rewritten test examples and applying similarity-based decontamination. Across multiple-choice Q&A, text classification, and open-ended Q&A with Qwen and Gemma models, BenchDoor achieves positive benchmark gains for all 30 evaluated configurations, with held-out attack success rates of 82.33% to 100.00% at poisoning rates of 2% to 10%. Both objectives remain achievable after synthetic data generation and similarity-based decontamination, with attack success rates of at least 90.33%. These findings show how benchmarks can serve as attack vectors both through their own data and through training resources that deliver improvements on them.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Benchmark (surveying), Backdoor, Exploit, Adversarial system, Computer science, Adversary

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange