analiza aiqbench
ForeSight addresses the critical challenge of early safety detection in large language models by distilling weak initial signals from first-token hidden states into compact, layer-aware risk representations. This approach overcomes limitations of existing methods that either rely on surface tokens or retain redundant information, offering a more efficient framework for forecasting harmful content generation. The research demonstrates superior performance across multiple benchmarks while operating solely on early post-generation states, marking a significant step forward in AI safety mechanisms.
ForeSight: Enhancing Risk Monitoring via Early Safety Signal Distillation
arXiv:2609.13737v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly deployed, the generation of harmful content has become a critical safety concern. Existing safeguards operate at the input, output, or streaming-generation stages, while early-risk methods that rely on surface tokens or output logits may suffer from weak initial signals, and internals-based detectors using dense representations may retain highly entangled and redundant safety-irrelevant information. It therefore remains unclear whether the earliest post-generation hidden states already contain reliable signals about final-response harmfulness. To address this gap, we propose ForeSight, a first-token output-risk forecasting framework that distills weak and redundant early safety signals into compact, layer-aware risk representations. Experiments on five safety benchmarks and two target models demonstrate that ForeSight achieves superior and efficient early-risk forecasting while relying solely on first-token hidden states. The code is available at: https://github.com/Scabbards1500/Foresight
Abstract:As large language models (LLMs) are increasingly deployed, the generation of harmful content has become a critical safety concern. Existing safeguards operate at the input, output, or streaming-generation stages, while early-risk methods that rely on surface tokens or output logits may suffer from weak initial signals, and internals-based detectors using dense representations may retain highly entangled and redundant safety-irrelevant information. It therefore remains unclear whether the earliest post-generation hidden states already contain reliable signals about final-response harmfulness. To address this gap, we propose ForeSight, a first-token output-risk forecasting framework that distills weak and redundant early safety signals into compact, layer-aware risk representations. Experiments on five safety benchmarks and two target models demonstrate that ForeSight achieves superior and efficient early-risk forecasting while relying solely on first-token hidden states. The code is available at: this https URL
17 pages, 12 figures. Accepted to Findings of EMNLP 2026
Computation and Language (cs.CL)
arXiv:2609.13737 [cs.CL]
(or arXiv:2609.13737v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.13737 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 240 cuvinte
Subiecte conexe
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · 5 oct. 2026
News
Quantum Computing Report · 7 iun. 2026
Digital States 2026: AI Moves From Experiment to Infrastructure
govtech.com · 24 sept. 2026
Against Export Controls (and China Threat Models)
LessWrong · 24 sept. 2026