[Submitted on 3 Jun 2026]

View PDF HTML (experimental)

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.

Submission history

From: Sanjay Kumar Mishra [view email]
[v1] Wed, 3 Jun 2026 02:55:17 UTC (921 KB)