# Quickest Detection of Hallucination Onset: Delay Bounds and Learned CUSUM Statistics

arXiv: https://arxiv.org/abs/2606.12476 · HTML: https://mlsanity.com/paper/2606.12476/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 12 of its 29 references that are in the map.
- First posted on arXiv 10 Jun 2026.

## Builds on (most cited first, 12 of 12)

- [SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models](https://mlsanity.com/paper/2303.08896/) (2023)
- [RAGTruth: A Hallucination Corpus for Developing Trustworthy Retrieval-Augmented Language Models](https://mlsanity.com/paper/2401.00396/) (2023)
- [A Token-level Reference-free Hallucination Detection Benchmark for Free-form Text Generation](https://mlsanity.com/paper/2104.08704/) (2021)
- [Real-Time Detection of Hallucinated Entities in Long-Form Generation](https://mlsanity.com/paper/2509.03531/) (2025)
- [HALT: Hallucination Assessment via Log-probs as Time series](https://mlsanity.com/paper/2602.02888/) (2026)
- [Neural network-based CUSUM for online change-point detection](https://mlsanity.com/paper/2210.17312/) (2022)
- [Recurrent Neural Networks and Universal Approximation of Bayesian Filters](https://mlsanity.com/paper/2211.00335/) (2022)
- [First Hallucination Tokens Are Different from Conditional Ones](https://mlsanity.com/paper/2507.20836/) (2025)
- [Hallucination as Trajectory Commitment: Causal Evidence for Asymmetric Attractor Dynamics in Transformer Generation](https://mlsanity.com/paper/2604.15400/) (2026)
- [Where Does Reasoning Break? Step-Level Hallucination Detection via Hidden-State Transport Geometry](https://mlsanity.com/paper/2605.13772/) (2026)
- [Sequential statistical inference for Large Language Models: Representation, validity, and monitoring](https://mlsanity.com/paper/2606.07624/) (2026)
- [Exactly Optimal Bayesian Quickest Change Detection for Hidden Markov Models](https://mlsanity.com/paper/2009.00150/) (2020)

## Built on by (most cited first, 1 of 1)

- [Delayed Verification Destabilizes Multi-Agent LLM Belief: Instability Thresholds and Optimal Corrector Placement](https://mlsanity.com/paper/2606.27409/) (2026)

## Abstract (arXiv)

Token-level hallucination detectors are evaluated as classifiers, by AUC over all tokens, yet a streaming monitor is judged by its reaction time: the number of tokens that pass between the onset of a hallucination and the alarm. We formulate hallucination onset detection as a quickest change detection problem. A first-order Markov model of the latent faithful/hallucinated state, validated on RAGTruth, places the task inside classical change-point theory and yields Lorden's lower bound on detection delay: about 1.3 tokens at a false-alarm rate of 0.01. We then show that a causal recurrent labeler acts as a CUSUM with a learned increment. Among the onsets it catches it detects in 11-13 tokens, against 31 for a linear per-token baseline, though at this false-alarm budget every detector catches under a third of onsets and the recall-honest delay is 56-66 tokens: low-false-alarm onset detection is hard. A controlled decomposition attributes the speed advantage mostly to a better per-token score rather than to temporal accumulation. An information-rate optimality theorem of Donsker-Varadhan type explains the remaining order-of-magnitude gap: the learned score realizes only 1/4.5 of the divergence the features carry, a deficit that recalibration cannot remove, with the remainder a finite-horizon effect. Classification metrics conceal this delay structure; sequential analysis makes it measurable.