# Kelp: A Streaming Safeguard for Large Models via Latent Dynamics-Guided Risk Detection

arXiv: https://arxiv.org/abs/2510.09694 · HTML: https://mlsanity.com/paper/2510.09694/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security.
- Builds on 40 of its 46 references that are in the map.
- First posted on arXiv 9 Oct 2025.

## Placed in

- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 30 of 40)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [GPT-4o System Card](https://mlsanity.com/paper/2410.21276/) (2024)
- [Neural Ordinary Differential Equations](https://mlsanity.com/paper/1806.07366/) (2018)
- [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://mlsanity.com/paper/2006.03654/) (2020)
- [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models](https://mlsanity.com/paper/2504.10479/) (2025)
- [Gemma 3 Technical Report](https://mlsanity.com/paper/2503.19786/) (2025)
- [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [Qwen2.5-Omni Technical Report](https://mlsanity.com/paper/2503.20215/) (2025)
- [DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models](https://mlsanity.com/paper/2306.11698/) (2023)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- [Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models](https://mlsanity.com/paper/2108.08877/) (2021)
- [FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts](https://mlsanity.com/paper/2311.05608/) (2023)
- [MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models](https://mlsanity.com/paper/2311.17600/) (2023)
- [PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference](https://mlsanity.com/paper/2406.15513/) (2024)
- [SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decoding](https://mlsanity.com/paper/2402.08983/) (2024)
- [JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks](https://mlsanity.com/paper/2404.03027/) (2024)
- [SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models](https://mlsanity.com/paper/2402.05044/) (2024)
- [Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails](https://mlsanity.com/paper/2501.09004/) (2025)
- [Llama Guard 3 Vision: Safeguarding Human-AI Image Understanding Conversations](https://mlsanity.com/paper/2411.10414/) (2024)
- [Closed-form Continuous-time Neural Models](https://mlsanity.com/paper/2106.13898/) (2021)

## Built on by (most cited first, 9 of 9)

- [The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook](https://mlsanity.com/paper/2604.02029/) (2026)
- [NExT-Guard: Training-Free Streaming Safeguard without Token-Level Labels](https://mlsanity.com/paper/2603.02219/) (2026)
- [Predict, Don't React: Value-Based Safety Forecasting for LLM Streaming](https://mlsanity.com/paper/2604.03962/) (2026)
- [SentGuard: Sentence-Level Streaming Guardrails for Large Language Models](https://mlsanity.com/paper/2606.02041/) (2026)
- [ROM: Real-time Overthinking Mitigation via Streaming Detection and Intervention](https://mlsanity.com/paper/2603.22016/) (2026)
- [AERIC: Anticipatory Hidden-State Monitoring for Implicit Harmful Dialogue](https://mlsanity.com/paper/2605.23974/) (2026)
- [SingProbe Technical Report](https://mlsanity.com/paper/2608.30703/) (2026)
- [Benchmarking Jailbreak Guardrails for Embodied Agents](https://mlsanity.com/paper/2610.06122/) (2026)
- [TRACE: Temporal Risk Assessment and Concept Editing for Safer Language Generation](https://mlsanity.com/paper/2610.15365/) (2026)

## Abstract (arXiv)

Large models (LMs) are powerful content generators, yet their open-ended nature can also introduce potential risks, such as generating harmful or biased content. Existing guardrails mostly perform post-hoc detection that may expose unsafe content before it is caught, and the latency constraints further push them toward lightweight models, limiting detection accuracy. In this work, we propose Kelp, a novel plug-in framework that enables streaming risk detection within the LM generation pipeline. Kelp leverages intermediate LM hidden states through a Streaming Latent Dynamics Head (SLD), which models the temporal evolution of risk across the generated sequence for more accurate real-time risk detection. To ensure reliable streaming moderation in real applications, we introduce an Anchored Temporal Consistency (ATC) loss to enforce monotonic harm predictions by embedding a benign-then-harmful temporal prior. Besides, for a rigorous evaluation of streaming guardrails, we also present StreamGuardBench-a model-grounded benchmark featuring on-the-fly responses from each protected model, reflecting real-world streaming scenarios in both text and vision-language tasks. Across diverse models and datasets, Kelp consistently outperforms state-of-the-art post-hoc guardrails and prior plug-in probes (15.61% higher average F1), while using only 20M parameters and adding less than 0.5 ms of per-token latency.