aiqbench analysis
This study introduces CyBiasBench, a new benchmark revealing that autonomous LLM agents in offensive cybersecurity scenarios exhibit stable selection biases regarding attack families and resist external steering attempts. The research demonstrates that forcing distribution shifts away from these inherent biases fails to improve attack performance, highlighting the need for better evaluation metrics beyond aggregate success rates in AI-driven security operations.
CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios
From arXiv
Full textarXiv:2605.07830v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly deployed as autonomous agents in offensive cybersecurity. In this paper, we reveal an interesting phenomenon: different agents exhibit distinct attack patterns. Specifically, agents exhibit distinct attack-selection biases, differing in both the families they explore and the concentration of their efforts. These patterns remain relatively stable across prompt variations. To systematically quantify this behavior, we introduce CyBiasBench, a comprehensive 630-session benchmark that evaluates five agents on three targets and four prompt conditions with ten attack families.
We identify explicit bias across agents, with different dominant attack families and varying entropy levels in their attack-family allocation distributions. Such bias characterizes agent behavior within the evaluated environments and provides information beyond aggregate attack success rates. Furthermore, our experiments reveal a bias momentum effect, where agents resist explicit steering toward attack families that conflict with their bias. This forced distribution shift does not yield measurable improvements in attack performance.
To ensure reproducibility and facilitate future research, we release an interactive result dashboard at https://trustworthyai.co.kr/CyBiasBench/ and a reproducibility artifact with aggregated session-level statistics and full evaluation scripts at https://github.com/Harry24k/CyBiasBench.
Abstract:Large language models (LLMs) are increasingly deployed as autonomous agents in offensive cybersecurity. In this paper, we reveal an interesting phenomenon: different agents exhibit distinct attack patterns. Specifically, agents exhibit distinct attack-selection biases, differing in both the families they explore and the concentration of their efforts. These patterns remain relatively stable across prompt variations. To systematically quantify this behavior, we introduce CyBiasBench, a comprehensive 630-session benchmark that evaluates five agents on three targets and four prompt conditions with ten attack families.
We identify explicit bias across agents, with different dominant attack families and varying entropy levels in their attack-family allocation distributions. Such bias characterizes agent behavior within the evaluated environments and provides information beyond aggregate attack success rates. Furthermore, our experiments reveal a bias momentum effect, where agents resist explicit steering toward attack families that conflict with their bias. This forced distribution shift does not yield measurable improvements in attack performance.
To ensure reproducibility and facilitate future research, we release an interactive result dashboard at this https URL and a reproducibility artifact with aggregated session-level statistics and full evaluation scripts at this https URL.
Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
arXiv:2605.07830 [cs.CR]
(or arXiv:2605.07830v2 [cs.CR] for this version)
https://doi.org/10.48550/arXiv.2605.07830 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.
Full text under the publisher's open licence · 280 words