analiza aiqbench
This research addresses a critical gap in LLM safety by introducing stateful guardrails that detect harm accumulating over multi-turn conversations rather than evaluating prompts in isolation. The authors propose a new framework tracking semantic drift and information accumulation, supported by extensive benchmarks designed to reduce common evaluation artifacts like template effects. This work is essential for deploying robust conversational AI systems where risk emerges gradually through dialogue dynamics.
Stateful Guardrails for Multi-Turn LLM Systems: A Conversational Risk Accumulation Framework
arXiv:2607.19361v1 Announce Type: cross Abstract: Most safety guardrails for large language models (LLMs) evaluate each prompt-response pair in isolation, which misses failures that arise only over a dialogue as benign turns compose into harm. We term this Conversational Risk Accumulation (CRA): gradual intent drift, fragmented assembly of prohibited instructions, and sensitivity build-up from repeated disclosures. We propose a session-layer CRA Framework that tracks three trajectory signals: semantic drift from a session anchor, a sensitivity-weighted information accumulation graph over extracted entities, and a compliance-gradient signal capturing increasing willingness to comply. For scoring, we provide (i) an unsupervised convex fusion for attribution and ablations, and (ii) CRA-Net DA, a compact learned trajectory model trained with family-adversarial objectives to reduce length and topic-coverage confounds. To benchmark CRA, we release CRA-Bench v0.1 (1,200 eight-turn sessions across three threat families with topic-matched benign twins), CRA-Bench v0.2 (LLM-paraphrased variants to reduce template artifacts), and an extended 5-family set (2,000 sessions adding persona priming and context stuffing). We introduce a trajectory-native evaluation protocol with session-level splits, mixed-set threshold calibration, Trajectory AUROC, turns-to-detection, calibrated false-positive metrics, bootstrap confidence intervals, leave-one-family-out diagnostic stress tests, and synthetic-to-human transfer checks. Claims focus on within-distribution session scoring on CRA-Bench and human-transfer subsets.
Abstract:Most safety guardrails for large language models (LLMs) evaluate each prompt-response pair in isolation, which misses failures that arise only over a dialogue as benign turns compose into harm. We term this Conversational Risk Accumulation (CRA): gradual intent drift, fragmented assembly of prohibited instructions, and sensitivity build-up from repeated disclosures. We propose a session-layer CRA Framework that tracks three trajectory signals: semantic drift from a session anchor, a sensitivity-weighted information accumulation graph over extracted entities, and a compliance-gradient signal capturing increasing willingness to comply. For scoring, we provide (i) an unsupervised convex fusion for attribution and ablations, and (ii) CRA-Net DA, a compact learned trajectory model trained with family-adversarial objectives to reduce length and topic-coverage confounds. To benchmark CRA, we release CRA-Bench v0.1 (1,200 eight-turn sessions across three threat families with topic-matched benign twins), CRA-Bench v0.2 (LLM-paraphrased variants to reduce template artifacts), and an extended 5-family set (2,000 sessions adding persona priming and context stuffing). We introduce a trajectory-native evaluation protocol with session-level splits, mixed-set threshold calibration, Trajectory AUROC, turns-to-detection, calibrated false-positive metrics, bootstrap confidence intervals, leave-one-family-out diagnostic stress tests, and synthetic-to-human transfer checks. Claims focus on within-distribution session scoring on CRA-Bench and human-transfer subsets.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
I.2.7; K.6.5; D.4.6
arXiv:2607.19361 [cs.CL]
(or arXiv:2607.19361v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2607.19361 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 · 296 cuvinte
Subiecte conexe
Performance Profiling on AMD GPUs – Part 5: Profiling-Driven Kernel Optimization with an AI Code-Assist Tool
AMD · 16 iul. 2026
Vertical Advantage: Transforming Industries with Lakebase and Agentic AI
Databricks · 27 aug. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026
Deploying Qwen3.8-2.4T-A95B on Amazon SageMaker HyperPod with vLLM
AWS · 9 sept. 2026