ai-safe2-framework

AI SAFE² v3.1 Surface: Research Context: v3.1 Current

Framework Home Research Index Cross-Pillar Governance AISM NEXUS Dashboard

Current framework context: AI SAFE² v3.1. This research note preserves its original publication date, evidence, and historical framework references. Use current v3.1 normative control and profile documents for implementation or conformance decisions.

Research Note: Indirect Prompt Injection via Memory (MINJA)

ID: RN-2025-005 | Related Control: [P1.T1.5_ADV], [P5.T1.4_ADV] | Status: Verified

🚨 The Threat Vector

Long-Term Memory Poisoning: Unlike immediate prompt injection, MINJA (Memory Injection) attacks target the agent’s long-term storage (Vector DB).

🛡️ The AI SAFE² Solution

We implement “Cognitive Hygiene” protocols for long-term storage.

1. Pre-Ingestion Sanitization [P1.T1.5_ADV]

All documents destined for RAG/Memory must be scrubbed of “Control Characters” and adversarial patterns before embedding.

2. Semantic Drift Detection [P4.T2.3_ADV]

Monitoring the vector space for “Clustering Anomalies.” If a new memory chunk sits in a semantic region known for jailbreaks, it is flagged before retrieval.

📚 References


Research navigation

Previous research note Research Index Next research note
Framework Home Cross-Pillar Governance NEXUS Challenge Lab

AI SAFE² v3.1 | Cyber Strategy Institute