Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 9 of the 15 references it cites were found in a registry or an index, and 1 are not checked yet.

PROMPT PERSISTENCE ATTACKS: LONG-TERM MEMORY POISONING IN LLM-BASED SYSTEMS

Open access
Bhatnagar, Pranav · 2026 · SSRN (preprint)

Abstract

Large Language Models (LLMs) are increasingly deployed as persistent, interactive systems that retain information across user interactions. These memory mechanisms are designed to enhance personalization, task continuity, and operational efficiency. However, persistence introduces a new and insufficiently examined security risk: the gradual corruption of long-term memory through legitimate interaction. This paper introduces the concept of Prompt Persistence Attacks, a class of long-horizon adversarial strategies that exploit memory retention in deployed LLMbased systems. Unlike prompt injection or jailbreak attacks that aim to induce immediate policy violations, prompt persistence attacks operate incrementally, shaping system memory over time in ways that remain below conventional detection thresholds. We analyze how contemporary memory architectures used in real-world LLM deployments can be manipulated without violating explicit safety constraints. Through detailed threat modeling and scenario-based analysis, we demonstrate how poisoned memory can bias future reasoning, distort decisionmaking, and undermine alignment guarantees. The paper concludes by outlining the limitations of existing defenses and identifying key research challenges in securing memory-enabled AI systems.

Overview

Integrity signals

Registry

Crossref · 10.2139/ssrn.6183548

Cites

9 of 15 references verified0 citation errors · 5 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Verification

Claims the paper makes

We couldn’t retrieve this paper’s full text, so its claims and self-check weren’t run. The registry and citation checks still apply.

Citations

Cites

Show all 15 references

Metadata

Authors
Bhatnagar, P.
Corresponding author
Bhatnagar, Pranav
Venue
SSRN (preprint)
Publisher
Elsevier BV
Published
2026
ISSN
1556-5068 (ISSN-L)
OpenAlex
W7127941464
Cited by
0
References
15
Type
posted content
Language
en
Open access
Locations
SSRN Electronic Journal (submittedVersion)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
  • Physical Sciences › Computer Science › Hardware and Architecture › Security and Verification in Computing
  • Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
Keywords
large language models, memory retention, adversarial attacks, AI alignment, threat modeling

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.