# Injection-Execution Dissociation: A Mechanistic Evaluation of Persistent Memory Attacks and Defenses in Stateful LLM Agents

arXiv: https://arxiv.org/abs/2605.08442 · HTML: https://mlsanity.com/paper/2605.08442/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 11 of its 21 references that are in the map.
- First posted on arXiv 8 May 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)

## Builds on (most cited first, 11 of 11)

- [Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection](https://mlsanity.com/paper/2302.12173/) (2023)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training](https://mlsanity.com/paper/2401.05566/) (2024)
- [Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents](https://mlsanity.com/paper/2410.02644/) (2024)
- [Formalizing and Benchmarking Prompt Injection Attacks and Defenses](https://mlsanity.com/paper/2310.12815/) (2023)
- [ShieldGemma: Generative AI Content Moderation Based on Gemma](https://mlsanity.com/paper/2407.21772/) (2024)
- [AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents](https://mlsanity.com/paper/2406.13352/) (2024)
- [Memory Injection Attacks on LLM Agents via Query-Only Interaction](https://mlsanity.com/paper/2503.03704/) (2025)
- [MemoryGraft: Persistent Compromise of LLM Agents via Poisoned Experience Retrieval](https://mlsanity.com/paper/2512.16962/) (2025)
- [Zombie Agents: Persistent Control of Self-Evolving LLM Agents via Self-Reinforcing Injections](https://mlsanity.com/paper/2602.15654/) (2026)
- [Trojan Hippo: Weaponizing Agent Memory for Data Exfiltration](https://mlsanity.com/paper/2605.01970/) (2026)

## Built on by (most cited first, 5 of 5)

- [MemLineage: Lineage-Guided Enforcement for LLM Agent Memory](https://mlsanity.com/paper/2605.14421/) (2026)
- [Online Shift Detection and Conformal Adaptation for Deployed Safety Classifiers](https://mlsanity.com/paper/2606.11949/) (2026)
- [Forensic Trajectory Signatures for Agent Memory Poisoning Detection](https://mlsanity.com/paper/2606.30566/) (2026)
- [Recognition Without Enforcement: Configuration-Dependent Failures in LLM Agent Instruction Arbitration and External Control](https://mlsanity.com/paper/2608.28502/) (2026)
- [From Attack Success to Attack Severity: Counterfactual Memory Attacks on LLM Agents](https://mlsanity.com/paper/2609.34132/) (2026)

## Abstract (arXiv)

We discover that prompt-injection success and tool-execution success are separable safety properties: defenses that block injection do not necessarily block execution, and vice versa. We call this the injection-execution dissociation. In LLM agents with persistent memory, malicious instructions are stored at rates exceeding 97.5%, yet downstream execution ranges from 0% to 95% with no correlation to storage rate. This reframes the threat model: preventing storage alone is insufficient, and blocking execution requires structurally enforcing authority boundaries between memory ingestion and action execution. We substantiate this through a 5,040-run factorial experiment across nine open-source models (N=40 per condition), evaluating six defenses at four architectural layers against delayed-trigger attacks that persist across session boundaries via RAG retrieval. Defense effectiveness is governed by where a defense sits relative to the attack's authority boundary, not by classifier quality. Only Memory Sandbox -- a tool-layer defense that structurally isolates recalled memory from executable context -- reduces attack success to 0% for eight of nine models. A reasoning-mode ablation reveals a double dissociation: no single schema-layer intervention is safe across both reasoning and non-reasoning model classes. A loaded-corpus frontier evaluation (21 models, 3 providers; N=40 base, headline models topped up to N=172) reveals vendor-correlated patterns: Anthropic blocks predominantly at injection, OpenAI blocks at execution with variable generational hardening, and a pre-release Gemini endpoint exfiltrates in the majority of runs. Stored-but-dormant payloads constitute a compositional supply-chain risk in shared-memory deployments.