analiza aiqbench
This research identifies "drunk language" as a novel attack vector that significantly increases LLM susceptibility to jailbreaking and privacy leaks compared to standard defenses. By simulating human intoxication through persona prompting and fine-tuning, the study reveals critical vulnerabilities in current safety tuning protocols for large models. The findings underscore the need for robustness testing against anthropomorphic behavioral shifts induced by specific linguistic patterns.
In Vino Veritas and Vulnerabilities: Examining LLM Safety via Drunk Language Inducement
arXiv:2601.22169v2 Announce Type: replace-cross Abstract: Humans are susceptible to undesirable behaviours and privacy leaks under the influence of alcohol. This paper investigates drunk language, i.e., text written under the influence of alcohol, as a driver for safety failures in large language models (LLMs). We investigate three mechanisms for inducing drunk language in LLMs: persona-based prompting, causal fine-tuning, and reinforcement-based post-training. When evaluated on 5 LLMs, we observe a higher susceptibility to jailbreaking on JailbreakBench (even in the presence of defences) and privacy leaks on ConfAIde, where both benchmarks are in English, as compared to the base LLMs as well as previously reported approaches. Via a robust combination of manual evaluation and LLM-based evaluators and analysis of error categories, our findings highlight a correspondence between human-intoxicated behaviour, and anthropomorphism in LLMs induced with drunk language. The simplicity and efficiency of our drunk language inducement approaches position them as potential counters for LLM safety tuning, highlighting significant risks to LLM safety.
Abstract:Humans are susceptible to undesirable behaviours and privacy leaks under the influence of alcohol. This paper investigates drunk language, i.e., text written under the influence of alcohol, as a driver for safety failures in large language models (LLMs). We investigate three mechanisms for inducing drunk language in LLMs: persona-based prompting, causal fine-tuning, and reinforcement-based post-training. When evaluated on 5 LLMs, we observe a higher susceptibility to jailbreaking on JailbreakBench (even in the presence of defences) and privacy leaks on ConfAIde, where both benchmarks are in English, as compared to the base LLMs as well as previously reported approaches. Via a robust combination of manual evaluation and LLM-based evaluators and analysis of error categories, our findings highlight a correspondence between human-intoxicated behaviour, and anthropomorphism in LLMs induced with drunk language. The simplicity and efficiency of our drunk language inducement approaches position them as potential counters for LLM safety tuning, highlighting significant risks to LLM safety.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
arXiv:2601.22169 [cs.CL]
(or arXiv:2601.22169v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2601.22169 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.
Text integral sub licența deschisă a publicației · 257 cuvinte
Subiecte conexe
Biological Superintelligence
LessWrong · 30 iul. 2026
Relevance as a Vulnerability: How Web Retrieval Degrades Safety Alignment in LLM Agents
arXiv · 1 sept. 2026
Minimal, Local, Causal Explanations for Jailbreak Success in Large Language Models
arXiv · 5 aug. 2026
An unexamined cause of the OpenAI Hugging Face hacking incident: its binary performance metric
LessWrong · 23 sept. 2026