LOTL Attacks Using Local LLMs
How future devices with built-in LLMs will become a security problem, because attackers will be able to live off the LLM (LOLLM)
Notes
Architecting secure enterprise AI agents with MCP
A guide to designing secure enterprise AI agents using MCP from IBM, with verification from Anthropic
Defending MLLMs from Implicit Jailbreak Attacks
A new class of attacks where text and image look safe separately, but their combination carries malicious meaning
Pruning-Activated Attack
Model pruning can be used by an attacker
External Data Extraction Attacks against RAG
The paper studies a new class of attacks against RAG-type systems
Fine-Tuning Jailbreaks
The paper discusses vulnerabilities in fine-tuning systems for large language models under conditions close to real-world operation
Airgeddon loves WiFi
Nobody likes wires; everyone loves Wi-Fi
Tool Tweak
An attack on tool selection in agentic systems
FuncPoison - Poisoned Library
A new attack based on poisoning a function library, which makes it possible to replace agent behavior without changing their models
Red Teaming LLM Agents with MCP
MCP is a critical and vulnerable point in the trust chain of LLM agents