Policy dependency / Stack layer
mcp-shell Ships Security Off in One Deploy Path and Bypassable in the Other (CVE-2026-55580/55581/55582)
GitHub Security Advisories
Stack layer / Threat pattern
Anthropic's own weekly sales digest runs on Claude Code plus a BigQuery MCP connector and nine hand-written content rules
Claude by Anthropic
Stack layer / Threat pattern
Microsoft's Agent Lightning v1.0.1 ships an agent skill whose job is optimizing other agents, installed through gh skill
GitHub
Stack layer / Threat pattern
Gradient open-sources a full reinforcement-learning loop for training tool-using research agents with GRPO
GitHub
Policy dependency / Stack layer
A flow-centric policy language cut confirmed agent compromise from 33% to 0% on AgentDojo while raising utility
arXiv 2608.22868
Policy dependency / Stack layer
SMITH Trains Tool Creation and Tool Use in One Policy; a 4B Qwen3 Beats an Untrained 30B Tool-Writer at 79.8 Macro Accuracy
arXiv 2608.24571
Stack layer / Threat pattern
Headless Claude Code now auto-continues responses cut off mid-stream, and subagents that hit maxTurns return partial output instead of looking finished
Claude Code changelog
Stack layer / Threat pattern
Only 4-16% of security rules written in CLAUDE.md have a matching Claude Code built-in control
arXiv