Helpful to a Fault: Measuring Illicit Assistance in Multi-Turn, Multilingual LLM Agents

Nivya Talokar, Ayush Kumar Tarun, Murari Mandal, Maksym Andriushchenko, Antoine Bosselut
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:118454-118485, 2026.

Abstract

LLM-based agents execute real-world workflows via tools. These affordances enable ill-intended adversaries to also use these agents to carry out complex misuse scenarios. Existing agent-misuse benchmarks largely test single-prompt instructions, leaving a gap in measuring how agents end up helping with harmful or illegal tasks over multiple turns. We introduce STING (Sequential Testing of Illicit N-step Goal execution), an automated red-teaming framework that constructs a step-by-step illicit plan grounded in a benign persona and iteratively probes a target agent with adaptive follow-ups, using judge agents to track phase completion. We further introduce an analysis framework that models multi-turn red-teaming as a time-to-first-jailbreak random variable, enabling analysis tools like discovery curves, hazard-ratio attribution by attack language, and a new metric: Restricted Mean Jailbreak Discovery. Across AgentHarm scenarios, STING yields substantially higher illicit-task completion than single-turn prompting and chat-oriented multi-turn baselines adapted to tool-using agents. In multilingual evaluations across six non-English settings, we find that attack success and illicit-task completion do not consistently increase in lower-resource languages, diverging from common chatbot findings. Overall, STING provides a practical way to evaluate and stress-test agent misuse in realistic deployment settings, where interactions are inherently multi-turn and often multilingual. Our code is available at https://github.com/epfl-nlp/helpful-to-a-fault.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-talokar26a, title = {Helpful to a Fault: Measuring Illicit Assistance in Multi-Turn, Multilingual {LLM} Agents}, author = {Talokar, Nivya and Tarun, Ayush Kumar and Mandal, Murari and Andriushchenko, Maksym and Bosselut, Antoine}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {118454--118485}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/talokar26a/talokar26a.pdf}, url = {https://proceedings.mlr.press/v306/talokar26a.html}, abstract = {LLM-based agents execute real-world workflows via tools. These affordances enable ill-intended adversaries to also use these agents to carry out complex misuse scenarios. Existing agent-misuse benchmarks largely test single-prompt instructions, leaving a gap in measuring how agents end up helping with harmful or illegal tasks over multiple turns. We introduce STING (Sequential Testing of Illicit N-step Goal execution), an automated red-teaming framework that constructs a step-by-step illicit plan grounded in a benign persona and iteratively probes a target agent with adaptive follow-ups, using judge agents to track phase completion. We further introduce an analysis framework that models multi-turn red-teaming as a time-to-first-jailbreak random variable, enabling analysis tools like discovery curves, hazard-ratio attribution by attack language, and a new metric: Restricted Mean Jailbreak Discovery. Across AgentHarm scenarios, STING yields substantially higher illicit-task completion than single-turn prompting and chat-oriented multi-turn baselines adapted to tool-using agents. In multilingual evaluations across six non-English settings, we find that attack success and illicit-task completion do not consistently increase in lower-resource languages, diverging from common chatbot findings. Overall, STING provides a practical way to evaluate and stress-test agent misuse in realistic deployment settings, where interactions are inherently multi-turn and often multilingual. Our code is available at https://github.com/epfl-nlp/helpful-to-a-fault.} }
Endnote
%0 Conference Paper %T Helpful to a Fault: Measuring Illicit Assistance in Multi-Turn, Multilingual LLM Agents %A Nivya Talokar %A Ayush Kumar Tarun %A Murari Mandal %A Maksym Andriushchenko %A Antoine Bosselut %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-talokar26a %I PMLR %P 118454--118485 %U https://proceedings.mlr.press/v306/talokar26a.html %V 306 %X LLM-based agents execute real-world workflows via tools. These affordances enable ill-intended adversaries to also use these agents to carry out complex misuse scenarios. Existing agent-misuse benchmarks largely test single-prompt instructions, leaving a gap in measuring how agents end up helping with harmful or illegal tasks over multiple turns. We introduce STING (Sequential Testing of Illicit N-step Goal execution), an automated red-teaming framework that constructs a step-by-step illicit plan grounded in a benign persona and iteratively probes a target agent with adaptive follow-ups, using judge agents to track phase completion. We further introduce an analysis framework that models multi-turn red-teaming as a time-to-first-jailbreak random variable, enabling analysis tools like discovery curves, hazard-ratio attribution by attack language, and a new metric: Restricted Mean Jailbreak Discovery. Across AgentHarm scenarios, STING yields substantially higher illicit-task completion than single-turn prompting and chat-oriented multi-turn baselines adapted to tool-using agents. In multilingual evaluations across six non-English settings, we find that attack success and illicit-task completion do not consistently increase in lower-resource languages, diverging from common chatbot findings. Overall, STING provides a practical way to evaluate and stress-test agent misuse in realistic deployment settings, where interactions are inherently multi-turn and often multilingual. Our code is available at https://github.com/epfl-nlp/helpful-to-a-fault.
APA
Talokar, N., Tarun, A.K., Mandal, M., Andriushchenko, M. & Bosselut, A.. (2026). Helpful to a Fault: Measuring Illicit Assistance in Multi-Turn, Multilingual LLM Agents. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:118454-118485 Available from https://proceedings.mlr.press/v306/talokar26a.html.

Related Material