AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

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

Bilgehan Sel, Xuanli He, Alwin Peng, Ming Jin, Jerry Wei

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add hybrid RL+SFT training after SFT-only stages (ablation)Trojan-Speak on Haiku 4.5 · Bug Bounty Benchmarkrubric score Avg@5 (%): SFT-only 55-65 -> 85.2Improved
Interleave SFT steps with RL instead of pure RL (ablation)Haiku 4.5 RL on cipher-encoded questions · Nemotron-MCQA encoded accuracyaccuracyImproved
Use minimal frequency-ordered substitution cipher instead of Base64 or ROT13 (ablation)Haiku 4.5 · GPQA-Diamondaccuracy (%): 41.4 (Base64) -> 62.6Improved
Use Base64 with case swap rather than minimal cipher (ablation)Haiku 4.5 · GPQA-Diamondaccuracy (%): 62.6 -> 35.7Hurt
Substitute frequency-ordered letters (7+) instead of random letters (ablation)substitution cipher · CBRN query classifier bypassbypass rate (%): consistent 99+% with 7+ substitutionsImproved
Use fewer substituted letters (ablation)cipher complexity · GPQA-Diamond (Qwen3 14B, 32B)capability retentionImproved
Vary template structure and chunk length (32 to 256) (ablation)Trojan-Speak on Qwen3 14B · GPQA-Diamondaccuracy: standard deviation 0.53%, max spread 1.5%No clear effect
Increase LoRA rank (8 to 128) (ablation)Trojan-Speak on Qwen3 8B/14B/32B · GPQA-Diamond during Stage 2accuracyImproved
Use smaller base model (Qwen3 8B/4B vs 14B+) (ablation)Trojan-Speak · GPQA-Diamondcapability retention (%): 89% (14B) -> 53% (8B), 55% (4B)Hurt
Apply Trojan-Speak fine-tuning to Haiku 4.5Claude Haiku 4.5 · GPQA-Diamondaccuracy (%): 65.7 -> 62.6Hurt
Train activation probe on cipher-trained model rather than base model (ablation)Softmax and SWiM probes · Ciphered harmful content detectionAUC: 0.27-0.46 -> 0.97-0.99Improved
Add probe-evasion training strategies to the attack (ablation)Trojan-Speak · Activation probe detectionprobe detection performanceNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy