Stack layer / Threat pattern
Study of 44 Developers: AI Code Assistants Barely Improve Secure API Usage
arXiv
Policy dependency / Update thread
NetInjectBench: Naive Network-Ops Agents Take Unsafe Actions 82.5% of the Time Under Prompt Injection
arXiv
Policy dependency / Follow-up thread
New Paper 'aiAuthZ' Moves Agent Authorization Off-Host, Reports 0% Residual Attack Across 15 LLMs
arXiv
Stack layer / Follow-up thread
'Coding Agents Think Ahead of Time': Latent Probes Predict Outcomes 25 Steps Early
arXiv / Hacker News
Stack layer / Follow-up thread
Self-Guided Test-Time Training: Let the Model Pick Its Own Evidence Spans for +15% Long-Context Accuracy
arXiv / HuggingFace Daily Papers
Stack layer / Threat pattern
Mako: A 'Self-Evolving Agentic OS' That Writes New Exploits Against Live Targets and Hot-Loads Them Into Itself
arXiv
Stack layer / Threat pattern
MCPZoo Study: Scanners Call 96.89% of MCP Servers 'Risky' — But Under Half of Those Alerts Are Real
arXiv
Update thread / Follow-up thread
MemGhost / 'Stealth Memory Injection': One Email Plants Persistent False Memories in Personal AI Agents
arXiv (corroborated by The Hacker News)