#security

try —

8 papers match

cs.SE2026

CoGate: Confidence-Gated Co-Decoding for Secure Code Generation

Minghao Hu, Lannan Luo, Allen Roush +1

The paper introduces CoGate, a method that uses the confidence of a security expert model to gate its influence during co-decoding for generating more secure code with large langua…

#code generation#large language models#security#co-decoding
cs.SE2026

Impossible to hide secret ...: Uncovering Security and Privacy Issues in LLM-native IDEs

Mostafijur Rahman Akhond, Md Afif Al Mamun, Gias Uddin +1

The paper investigates security and privacy concerns reported by developers using LLM-native IDEs, analyzing Reddit discussions to build a taxonomy of issues and suggesting design…

#security#privacy#large language models#integrated development environments
cs.CR2026

AgentWorm: Self-Propagating Attacks Across LLM Agent Ecosystems

Yihao Zhang, Zeming Wei, Xiaokun Luan +7

The paper introduces AgentWorm, a self-replicating worm that can autonomously infect and spread across large-scale LLM-based agent ecosystems by hijacking configurations and execut…

#llm agents#self-replicating worm#agent ecosystems#security
cs.NI2026

Design and Benchmarking of the Data Distribution Service for Real-Time Interoperable Agricultural Machinery Communications

Samuel Brodie, Henri Hornburg, Daniel Ostermeier +2

The paper evaluates the Data Distribution Service (DDS) as middleware for secure, plug‑and‑play Ethernet‑based communication between agricultural machines, presenting a C++ proof‑o…

#agricultural machinery communication#data distribution service#middleware#real-time networking
cs.CR2026

Bad Memory: Evaluating Prompt Injection Risks from Memory in Agentic Systems

Soham Gadgil, David Alexander, Sai Sunku +1

The paper investigates how malicious instructions embedded in persistent memory files can be used to launch prompt injection attacks on agentic AI systems, evaluating several large…

#prompt injection#agentic systems#persistent memory#security
cs.DC2026

Memory-Exhaustion Attack on the Blocklace Byzantine-Repelling Conflict-Free Replicated Data Type

Erick Lavoie

The paper demonstrates that the blocklace CRDT’s replication rules can be exploited to cause a memory‑exhaustion attack, allowing malicious nodes to flood correct nodes with arbitr…

#byzantine fault tolerance#conflict-free replicated data type#blocklace#memory exhaustion attack