EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Reflect-Guard: Enhancing LLM Safeguards Against Adversarial Prompts via Logical Self-Reflection

Lixing Lin, Juli You, Yue Li, Luyun Lin, Yiqing Wang, Zhen Zhang, Moxuan Zheng

Journal of Superintelligence · 2026

DOI: 10.3390/superintelligence1020007 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Superintelligence, 2026. Verfasst von Lixing Lin und 6 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

LLM safety classifiers such as Llama Guard detect overtly harmful prompts well but remain vulnerable to jailbreaks that disguise intent via role-play, fictional framing, or indirect requests. We present Reflect-Guard: Llama-Guard-3-8B fine-tuned via QLoRA to generate a distilled, GPT-4o-mini-sourced safety rationale before its verdict. Using only 1000training examples and a five-condition ablation, we isolate reflection’s contribution from that of ordinary supervised fine-tuning (SFT). SFT on verdict labels alone already accounts for most of the aggregate gain over the un-fine-tuned baseline (WildGuardTest F1: 0.770 → 0.860; JailbreakBench guard bypass rate: 10.3% → 0.0%). Adding reflection does not improve these aggregate numbers further (F1 falls to 0.842, guard bypass rises to 1.8%) but reallocates performance toward adversarially framed inputs: adversarial-subset recall rises from 0.845 (SFT-only) to 0.921 (+7.6 pp), at a cost of −7.9 pp precision on that subset. The same direction holds across three training seeds (paired t-tests, p=0.018 for overall F1 and p=0.027 for adversarial recall), though we interpret these values cautiously given the small number of seeds. Measured inference latency is 17.9× the baseline’s, driven by ∼92 additional reflection tokens per prompt. The same checkpoint, without additional fine-tuning, shows promising zero-shot transfer to a 329-example financial-harm benchmark (F1: 0.868 → 0.929; recall: 0.767 → 0.867, zero false positives); preliminary two-seed evidence, however, suggests the SFT-only condition may match or exceed this result, so we treat the finance improvement as evidence toward generalization rather than a settled advantage of reflection specifically. Benchmarked against WildGuard, PromptGuard, and a prompted (non-fine-tuned) GPT-4o-mini reasoning classifier on the same sets, WildGuard leads on aggregate WildGuardTest F1 (0.887), while Reflect-Guard achieves higher adversarial-subset recall than every evaluated classifier, including WildGuard (0.921 vs. 0.827) and the prompted teacher (0.636). Overall, supervised fine-tuning drives most of the aggregate gain over an un-fine-tuned classifier, while the generated safety rationale yields a narrower, latency-costly improvement specifically in adversarial-subset recall, a precision–recall tradeoff relevant to safety-critical deployments that prioritize recall over inference speed.

Zitieren

APA

Lin, L., You, J., Li, Y., Lin, L., Wang, Y., Zhang, Z., & Zheng, M. (2026). Reflect-Guard: Enhancing LLM Safeguards Against Adversarial Prompts via Logical Self-Reflection. Journal of Superintelligence. https://doi.org/10.3390/superintelligence1020007

Vancouver

Lin L, You J, Li Y, Lin L, Wang Y, Zhang Z, et al. Reflect-Guard: Enhancing LLM Safeguards Against Adversarial Prompts via Logical Self-Reflection. Journal of Superintelligence. 2026. doi:10.3390/superintelligence1020007

BibTeX
@article{Lin2026,
  author = {Lin, Lixing and You, Juli and Li, Yue and Lin, Luyun and Wang, Yiqing and Zhang, Zhen and Zheng, Moxuan},
  title = {Reflect-Guard: Enhancing LLM Safeguards Against Adversarial Prompts via Logical Self-Reflection},
  journal = {Journal of Superintelligence},
  year = {2026},
  doi = {10.3390/superintelligence1020007}
}
Kostenlos starten →