Atlas · GenAI 2026
AI Supply Chain Security
Supply chain security (models, dependencies, poisoning)
conceptPeak: 2024AI SecurityAI consensus: 2/3
Prerequisites
- mediumData Curation
Data poisoning defense requires understanding how training data is curated and what could be injected
Recommended reference
Carlini et al. (2024) 'Poisoning Web-Scale Training Datasets is Practical' — IEEE S&P; demonstrates real-world data poisoning attacks
Notes from AI deep research
Anthropic Opus
Carlini (2024): data poisoning jest praktycznie wykonalny. Modele i dane = wektory ataku
OpenAI Deep Research
Zależności jako wektor ataku [OA#71]
Google Deep Think
Skanowanie PDF zasilających RAG [G#86]
Related skills
- → is subcategory of: AI Risk Management(3/3)