Sari la conținut

Română · beta. Traducerea a fost făcută automat și este încă în curs de revizuire. Citește pagina în engleză

perspectiva aiqbenchSiguranță AI

analiza aiqbench

This research exposes a critical vulnerability in current AI safety evaluations, demonstrating that jailbreak robustness can fluctuate dramatically based on minor, non-safety-related changes to system prompts or operational states. The study reveals that relying solely on default configurations masks significant risks, as attack success rates can increase by over 50 percentage points without any modification to the attack vector itself. Consequently, the field must adopt more comprehensive testing protocols that account for these hidden state dependencies to ensure reliable alignment across real-world usage scenarios.
Surse care acoperăarXiv, LessWrong
Ce este aceastaLectura unui model asupra articolului și a acoperirii conexe, scrisă pentru aiqbench. Nu sunt cuvintele publicației.

The Fragility of Jailbreak Robustness Across Operational States

arXiv2 min de cititArticolul și perspectiva sunt în engleză.

arXiv:2608.30748v1 Announce Type: cross Abstract: Existing jailbreak evaluations typically characterize robustness using a single attack success rate (ASR) measured in a default configuration (the vanilla state). However, user-LLM interactions can induce diverse operational states beyond the vanilla state. In this work, we find that jailbreak robustness is highly fragile to operational-state variation: even when the attack remains fixed, changing only an ordinary system prompt not designed to affect safety can dramatically alter attack success rates. We systematically investigate this phenomenon across seven aligned models and three representative jailbreak attacks, observing substantial differences in ASR between vanilla and non-vanilla operational states. In one case, ASR increases by up to 56 percentage points (2% to 58%) solely due to a change in operational state. Remarkably, these increases occur even for attacks originally designed and optimized under vanilla-state evaluation. We further show that state-dependent robustness variation is systematically associated with differences in hidden representations along a refusal-related axis, and that projections onto this axis strongly predict jailbreak outcomes. Our results show that a single vanilla-state evaluation may not fully characterize jailbreak robustness, motivating evaluations that also examine how robustness changes across non-vanilla operational states.

Abstract:Existing jailbreak evaluations typically characterize robustness using a single attack success rate (ASR) measured in a default configuration (the vanilla state). However, user-LLM interactions can induce diverse operational states beyond the vanilla state. In this work, we find that jailbreak robustness is highly fragile to operational-state variation: even when the attack remains fixed, changing only an ordinary system prompt not designed to affect safety can dramatically alter attack success rates. We systematically investigate this phenomenon across seven aligned models and three representative jailbreak attacks, observing substantial differences in ASR between vanilla and non-vanilla operational states. In one case, ASR increases by up to 56 percentage points (2% to 58%) solely due to a change in operational state. Remarkably, these increases occur even for attacks originally designed and optimized under vanilla-state evaluation. We further show that state-dependent robustness variation is systematically associated with differences in hidden representations along a refusal-related axis, and that projections onto this axis strongly predict jailbreak outcomes. Our results show that a single vanilla-state evaluation may not fully characterize jailbreak robustness, motivating evaluations that also examine how robustness changes across non-vanilla operational states.

Cryptography and Security (cs.CR); Computation and Language (cs.CL)

arXiv:2608.30748 [cs.CR]

(or arXiv:2608.30748v1 [cs.CR] for this version)

https://doi.org/10.48550/arXiv.2608.30748 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

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.

Text integral sub licența deschisă a publicației · 285 cuvinte

Subiecte conexe