Trojan-Speak: Bypassing Constitutional Classifiers with No Jailbreak Tax via Adversarial Finetuning

Bilgehan Sel, Xuanli He, Alwin Peng, Ming Jin, Jerry Wei
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:108843-108861, 2026.

Abstract

Fine-tuning APIs offered by major AI providers create new attack surfaces where adversaries can bypass safety measures through targeted fine-tuning. We introduce Trojan-Speak, an adversarial fine-tuning method that bypasses Anthropic’s Constitutional Classifiers. Our approach uses curriculum learning combined with GRPO-based hybrid reinforcement learning to teach models a communication protocol that evades LLM-based content classification. Crucially, while prior adversarial fine-tuning approaches report more than 25% capability degradation on reasoning benchmarks, Trojan-Speak incurs less than 5% degradation while achieving 99+% classifier evasion for models with 14B+ parameters. We demonstrate that fine-tuned models can provide detailed responses to expert-level CBRN (Chemical, Biological, Radiological, and Nuclear) queries from Anthropic’s Constitutional Classifiers bug-bounty program. Our findings reveal that LLM-based content classifiers alone are insufficient for preventing dangerous information disclosure when adversaries have fine-tuning access, and we show that activation-level probes can substantially improve robustness to such attacks.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-sel26a, title = {Trojan-Speak: Bypassing Constitutional Classifiers with No Jailbreak Tax via Adversarial Finetuning}, author = {Sel, Bilgehan and He, Xuanli and Peng, Alwin and Jin, Ming and Wei, Jerry}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {108843--108861}, 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/sel26a/sel26a.pdf}, url = {https://proceedings.mlr.press/v306/sel26a.html}, abstract = {Fine-tuning APIs offered by major AI providers create new attack surfaces where adversaries can bypass safety measures through targeted fine-tuning. We introduce Trojan-Speak, an adversarial fine-tuning method that bypasses Anthropic’s Constitutional Classifiers. Our approach uses curriculum learning combined with GRPO-based hybrid reinforcement learning to teach models a communication protocol that evades LLM-based content classification. Crucially, while prior adversarial fine-tuning approaches report more than 25% capability degradation on reasoning benchmarks, Trojan-Speak incurs less than 5% degradation while achieving 99+% classifier evasion for models with 14B+ parameters. We demonstrate that fine-tuned models can provide detailed responses to expert-level CBRN (Chemical, Biological, Radiological, and Nuclear) queries from Anthropic’s Constitutional Classifiers bug-bounty program. Our findings reveal that LLM-based content classifiers alone are insufficient for preventing dangerous information disclosure when adversaries have fine-tuning access, and we show that activation-level probes can substantially improve robustness to such attacks.} }
Endnote
%0 Conference Paper %T Trojan-Speak: Bypassing Constitutional Classifiers with No Jailbreak Tax via Adversarial Finetuning %A Bilgehan Sel %A Xuanli He %A Alwin Peng %A Ming Jin %A Jerry Wei %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-sel26a %I PMLR %P 108843--108861 %U https://proceedings.mlr.press/v306/sel26a.html %V 306 %X Fine-tuning APIs offered by major AI providers create new attack surfaces where adversaries can bypass safety measures through targeted fine-tuning. We introduce Trojan-Speak, an adversarial fine-tuning method that bypasses Anthropic’s Constitutional Classifiers. Our approach uses curriculum learning combined with GRPO-based hybrid reinforcement learning to teach models a communication protocol that evades LLM-based content classification. Crucially, while prior adversarial fine-tuning approaches report more than 25% capability degradation on reasoning benchmarks, Trojan-Speak incurs less than 5% degradation while achieving 99+% classifier evasion for models with 14B+ parameters. We demonstrate that fine-tuned models can provide detailed responses to expert-level CBRN (Chemical, Biological, Radiological, and Nuclear) queries from Anthropic’s Constitutional Classifiers bug-bounty program. Our findings reveal that LLM-based content classifiers alone are insufficient for preventing dangerous information disclosure when adversaries have fine-tuning access, and we show that activation-level probes can substantially improve robustness to such attacks.
APA
Sel, B., He, X., Peng, A., Jin, M. & Wei, J.. (2026). Trojan-Speak: Bypassing Constitutional Classifiers with No Jailbreak Tax via Adversarial Finetuning. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:108843-108861 Available from https://proceedings.mlr.press/v306/sel26a.html.

Related Material