Red Specter LAZARUS

AI Memory Persistence Testing Framework — 7 subsystems. 96 tests.

v1.0.0

Contents

Overview Installation PLANTER — Memory Instruction Planting DORMANT — Dormant Trigger Testing WORM — Cross-Agent Propagation QUARANTINE — Quarantine Evasion RESURRECTOR — Persistence After Cleanup PROFILER — Memory Store Profiling SCANNER — Full Memory Scan UNLEASHED Mode CLI Reference MITRE ATLAS Mapping Disclaimer

Overview

LAZARUS targets AI memory systems — the persistent storage mechanisms that AI agents use to maintain context, remember instructions, and store user preferences across sessions. Every memory store is an attack surface. LAZARUS tests what survives deletion.

Deleted doesn't mean gone. LAZARUS tests what survives.

Installation

$ pip install red-specter-lazarus
$ lazarus init
$ lazarus status

PLANTER — Memory Instruction Planting

ID	Technique	Description
PL-001	Instruction Injection	Plant persistent instructions in AI memory stores
PL-002	Preference Manipulation	Alter stored user preferences to influence agent behaviour
PL-003	Context Poisoning	Poison stored context to redirect future interactions
PL-004	Cross-Session Planting	Plant instructions that persist across session boundaries

DORMANT — Dormant Trigger Testing

Test dormant triggers planted in AI memory. Time-based activation mechanisms. Context-triggered payload execution. Conditional instruction firing based on specific user inputs. Sleeper instructions that remainThis request triggered restrictions on violative cyber content and was blocked under Anthropic's Usage Policy. To learn more visit our help center: https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals