Latent Undertow: How Ordinary Typos Break Probes

Abstract:LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model's response substantively unchanged. Yet probes that detect malicious prompts by reading the model's hidden states tell a different story: the same edit rotates the readout vector by 43--56 at the perturbed token, decaying below 15% within ~10 downstream tokens. Stacking ~3 common typos per message cuts a single-position prompt-injection probe's TPR@FPR$=1% by 12.0pp, a gap recalibration alone cannot close. Multi-position aggregation cures localized perturbations (<= 0.5 loss) but only attenuates distributed ones, where even attention- and max-based aggregators still drop ~3.8pp. For single-position probes, we introduce a KV-cache fork: a short fixed suffix appended after the user message lets the probe read a few tokens downstream of the perturbation, exploiting its rapid spatial decay. This closes 95% of the gap (-0.6pp residual) -- an order of magnitude better than perturbation-augmented training (-3.7pp). The rotation-and-decay geometry replicates on Llama-3.1-8B, Qwen3-8B, and Gemma-4-E4B; probe evaluation is on Llama-3.1-8B. Code: this https URL

Computation and Language (cs.CL); Machine Learning (cs.LG)

arXiv:2609.15994 [cs.CL]

(or arXiv:2609.15994v1 [cs.CL] for this version)

https://doi.org/10.48550/arXiv.2609.15994 Focus to learn more arXiv-issued DOI via DataCite

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.