RAG Knowledge Base Poisoning: The Model Isn't Wrong — The Material It Reads Is.
RAG can make the model’s answers appear more evidence-based, but once the knowledge base is contaminated, the evidence itself becomes an attack surface....
Buno.dev
Things I build, study, and keep.
Recent
Prompt Injection: Do Not Trust the Context Read by the Model
Prompt injection is no simple prompt prank. It emerges when input, permission and execution boundaries are intertwined in LLM applications, posing genuine secur...
TRC20-USDT On-Chain Evidence Preservation Experiment
TRC20-USDT transaction records are publicly queryable, have relatively stable field structures, and can be cross-checked through multiple on-chain data interfac...