analiza aiqbench
This research introduces a training-free method for detecting policy violations in LLMs by analyzing normalized hidden activations, effectively framing the task as an out-of-distribution problem. The approach significantly reduces computational overhead compared to fine-tuning or LLM-as-a-judge baselines while achieving superior F1 scores on complex benchmarks. By leveraging internal model representations without retraining, this technique offers a lightweight solution for deploying aligned models in sensitive sectors like finance and law.
Training-Free Policy Violation Detection via Activation-Space Whitening in LLMs
arXiv:2512.03994v4 Announce Type: replace Abstract: As organizations increasingly deploy LLMs in sensitive domains such as legal, financial, and medical settings, ensuring alignment with internal organizational policies has become a priority. Existing content moderation frameworks remain largely confined to the safety domain and lack the robustness to capture nuanced organizational policies. LLM-as-a-judge and fine-tuning approaches, though flexible, introduce significant latency and training cost. To address these limitations, we frame policy violation detection as an out-of-distribution (OOD) problem in the model's activation space. We propose a training-free method that operates directly on the LLM internal representations, leveraging prior evidence that decision-relevant information is encoded within them. Inspired by whitening techniques, we derive policy-violation scores directly from normalized representations of LLM hidden activations. Our method requires only the policy text and a small number of illustrative samples, making it lightweight and easily deployable. We extensively evaluate our method across multiple LLMs and challenging policy benchmarks. It achieves up to 86.0% F1, outperforming both fine-tuned and LLM-as-a-judge baselines while requiring substantially less computation. Our code is publicly available at: https://github.com/FujitsuResearch/LLM-policy-violation-detection
Abstract:As organizations increasingly deploy LLMs in sensitive domains such as legal, financial, and medical settings, ensuring alignment with internal organizational policies has become a priority. Existing content moderation frameworks remain largely confined to the safety domain and lack the robustness to capture nuanced organizational policies. LLM-as-a-judge and fine-tuning approaches, though flexible, introduce significant latency and training cost. To address these limitations, we frame policy violation detection as an out-of-distribution (OOD) problem in the model's activation space. We propose a training-free method that operates directly on the LLM internal representations, leveraging prior evidence that decision-relevant information is encoded within them. Inspired by whitening techniques, we derive policy-violation scores directly from normalized representations of LLM hidden activations. Our method requires only the policy text and a small number of illustrative samples, making it lightweight and easily deployable. We extensively evaluate our method across multiple LLMs and challenging policy benchmarks. It achieves up to 86.0% F1, outperforming both fine-tuned and LLM-as-a-judge baselines while requiring substantially less computation. Our code is publicly available at: this https URL
Machine Learning (cs.LG)
arXiv:2512.03994 [cs.LG]
(or arXiv:2512.03994v4 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2512.03994 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 265 cuvinte
Subiecte conexe
[AINews] OpenAI DevDay 2026: Dots, 6.1 Sol, Ultrafast, Decisions API, Agents API, Spaces, Marketplace, and 1.2 Billion ChatGPT WAU
Latent Space · 30 sept. 2026
BRICS Draws a New Global Line: Full Text of New Delhi Declaration
slguardian.org · 12 sept. 2026
OpenRouter: from Seed to Stripe — with OpenRouter’s Alex Atallah & AMP’s Anjney Midha
Latent Space · 25 sept. 2026
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026