Atlas · GenAI 2026
AI Data Security
Data exfiltration defense in RAG/agent systems
conceptPeak: 2024AI SecurityAI consensus: 1/3
Prerequisites
- hardSecure RAG
Data exfiltration defense includes securing RAG retrieval — secure RAG is one component of the broader defense
Data exfiltration often exploits prompt injection to bypass access controls
Recommended reference
OWASP (2025) 'LLM Top 10: LLM06 Sensitive Information Disclosure' — covers exfiltration vectors specific to RAG/agent systems
Notes from AI deep research
Anthropic Opus
Exfiltration w RAG/agent. Kontrola kontekstu, polityki narzedzi, testy ujawniania
OpenAI Deep Research
Kontrola kontekstu, polityki narzędzi [OA#70]
Related skills
- → is subcategory of: AI Risk Management(3/3)
- ← is part of: PII Management(3/3)
- ← is part of: Prompt Injection Defense(3/3)
- ← is part of: AI Rate Limiting(2/3)
- ← is part of: Secure RAG(2/3)