# Peak + Accumulation: A Proxy-Level Scoring Formula for Multi-Turn LLM Attack Detection

arXiv: https://arxiv.org/abs/2602.11247 · HTML: https://mlsanity.com/paper/2602.11247/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Jailbreaking, one of 12 directions in LLM safety and security.
- Builds on 1 of its 15 references that are in the map.
- First posted on arXiv 11 Feb 2026.

## Placed in

- LLM safety and security / [Jailbreaking](https://mlsanity.com/safety/jailbreaking/)

## Builds on (most cited first, 1 of 1)

- [Multi-Turn Jailbreaks Are Simpler Than They Seem](https://mlsanity.com/paper/2508.07646/) (2025)

## Built on by (most cited first, 2 of 2)

- [The Workload-Router-Pool Architecture for LLM Inference Optimization: A Vision Paper from the vLLM Semantic Router Project](https://mlsanity.com/paper/2603.21354/) (2026)
- [Cognitive Firewall: A Proactive, Zero-Trust, Multi-Gate Framework for LLM Safety](https://mlsanity.com/paper/2607.01277/) (2026)

## Abstract (arXiv)

Multi-turn prompt injection attacks distribute malicious intent across multiple conversation turns, exploiting the assumption that each turn is evaluated independently. While single-turn detection has been extensively studied, no published formula exists for aggregating per-turn pattern scores into a conversation-level risk score at the proxy layer -- without invoking an LLM. We identify a fundamental flaw in the intuitive weighted-average approach: it converges to the per-turn score regardless of turn count, meaning a 20-turn persistent attack scores identically to a single suspicious turn. Drawing on analogies from change-point detection (CUSUM), Bayesian belief updating, and security risk-based alerting, we propose peak + accumulation scoring -- a formula combining peak single-turn risk, persistence ratio, and category diversity. Evaluated on 10,654 multi-turn conversations -- 588 attacks sourced from WildJailbreak adversarial prompts and 10,066 benign conversations from WildChat -- the formula achieves 90.8% recall at 1.20% false positive rate with an F1 of 85.9%. A sensitivity analysis over the persistence parameter reveals a phase transition at rho ~ 0.4, where recall jumps 12 percentage points with negligible FPR increase. We release the scoring algorithm, pattern library, and evaluation harness as open source.