Notes

Model pruning can be used by an attacker

The paper studies a new class of attacks against RAG-type systems

The paper discusses vulnerabilities in fine-tuning systems for large language models under conditions close to real-world operation

Nobody likes wires; everyone loves Wi-Fi

An attack on tool selection in agentic systems

A new attack based on poisoning a function library, which makes it possible to replace agent behavior without changing their models

MCP is a critical and vulnerable point in the trust chain of LLM agents

All you need is fuzzing…

A protocol that allows users to authenticate on a network and access services after authentication

NTLM is a Microsoft authentication protocol