When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift

arXiv:2602.14161v2 Announce Type: replace Abstract: Detecting prompt injection, jailbreak attacks, and harmful requests is critical for deploying LLM-based agents safely, yet current evaluation practices in this literature overestimate generalization. We train activation-based classifiers (linear probes on LLM hidden states) on a benchmark of 18 datasets (prompt attacks plus benign sources) and propose Leave-One-Dataset-Out (LODO) evaluation, where the held-out dataset is never seen during training. Across four LLMs from three families (Llama-3.1-8B, Gemma-3-27B, Qwen-3.5-2B/4B), standard cross-validation reports a pooled AUC 8.0-16.5 points higher than LODO; per-dataset held-out-test-vs-LODO accuracy gaps span 1-25 points. To understand the gap, we analyze the LODO stability of a linear probe's per-feature classifier coefficients, defining a retention metric for sparse-autoencoder (SAE) features that flags dataset-dependent shortcuts. 28-44% of top SAE features are shortcuts across models, a dataset-identity classifier reaches 96.6%, and the dataset-identifying and safety-relevant subspaces partially overlap. Standard domain-generalization fixes such as adversarial training, subspace projection, sample reweighting, and class balancing do not close the gap. Finally, we show LODO-weighted SAE attributions filter dataset artifacts for more reliable per-prompt explanations. We release our framework at https://github.com/maxf-zn/prompt-mining so future prompt-attack classifiers can be evaluated under LODO alongside CV.

Abstract:Detecting prompt injection, jailbreak attacks, and harmful requests is critical for deploying LLM-based agents safely, yet current evaluation practices in this literature overestimate generalization. We train activation-based classifiers (linear probes on LLM hidden states) on a benchmark of 18 datasets (prompt attacks plus benign sources) and propose Leave-One-Dataset-Out (LODO) evaluation, where the held-out dataset is never seen during training. Across four LLMs from three families (Llama-3.1-8B, Gemma-3-27B, Qwen-3.5-2B/4B), standard cross-validation reports a pooled AUC 8.0-16.5 points higher than LODO; per-dataset held-out-test-vs-LODO accuracy gaps span 1-25 points. To understand the gap, we analyze the LODO stability of a linear probe's per-feature classifier coefficients, defining a retention metric for sparse-autoencoder (SAE) features that flags dataset-dependent shortcuts. 28-44% of top SAE features are shortcuts across models, a dataset-identity classifier reaches 96.6%, and the dataset-identifying and safety-relevant subspaces partially overlap. Standard domain-generalization fixes such as adversarial training, subspace projection, sample reweighting, and class balancing do not close the gap. Finally, we show LODO-weighted SAE attributions filter dataset artifacts for more reliable per-prompt explanations. We release our framework at this https URL so future prompt-attack classifiers can be evaluated under LODO alongside CV.

Machine Learning (cs.LG)

arXiv:2602.14161 [cs.LG]

(or arXiv:2602.14161v2 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2602.14161 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

analiza aiqbench

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift

Subiecte conexe