0

Adversarial Manipulation of Reasoning Models using Internal Representations

Reasoning models are vulnerable to jailbreak attacks through manipulation of the chain-of-thought generation process, specifically by targeting a "caution" direction in activation space.

Year
2025
Venue
arXiv 2025
Authors
3
Hosting
Abstract onlyARXIV-DEFAULT

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text
arxiv.org/abs/2507.03167ARXIV-DEFAULT
TL;DR
Semantic Scholar
Attribution policy →

Abstract

Reasoning models generate chain-of-thought (CoT) tokens before their final output, but how this affects their vulnerability to jailbreak attacks remains unclear. While traditional language models make refusal decisions at the prompt-response boundary, we find evidence that DeepSeek-R1-Distill-Llama-8B makes these decisions within its CoT generation. We identify a linear direction in activation space during CoT token generation that predicts whether the model will refuse or comply -- termed the "caution" direction because it corresponds to cautious reasoning patterns in the generated text. Ablating this direction from model activations increases harmful compliance, effectively jailbreaking the model. We additionally show that intervening only on CoT token activations suffices to control final outputs, and that incorporating this direction into prompt-based attacks improves success rates. Our findings suggest that the chain-of-thought itself is a promising new target for adversarial manipulation in reasoning models. Code available at https://github.com/ky295/reasoning-manipulation

Authors

3