Policy dependency / Stack layer
Claude adds skill and plugin security scanning for Enterprise the same day Agent Plugins ships
Anthropic Release Notes
Policy dependency / Stack layer
AI Explained Stitches the Week Together: 10 Autonomous Math Discoveries, the Agent Message Board, a 'Constitutional Failure' and Google's Leadership Blowup
AI Explained (YouTube)
Policy dependency / Stack layer
Enterprise coding agents execute malicious skill files in 95.5-96.1% of runs — and notice something is wrong only 1.99% of the time
arXiv 2608.05223
Policy dependency / Stack layer
GitHub's lawyers built their own Copilot CLI agents in Markdown and halved contract review time
GitHub Blog
Policy dependency / Stack layer
Item Response Theory Fit to 8 Safety Benchmarks Across 192 Models Cuts Evaluation Cost 97-99% and Detects Naive Sandbagging
arXiv 2608.05086
Policy dependency / Stack layer
Agent Signing Keys Move Into HSMs: PKCS#11 Keystore Plus Zero-Trust MCP Stack Drops Injection Success From 19.3% to 0%
arXiv 2608.06130
Policy dependency / Stack layer
The Verge Traces 'Spiralism,' an AI-Originated Belief System That Chatbots Started and Humans Joined
The Verge
Policy dependency / Stack layer
A New 'Agent Runtime Security' Product Category Materialized at Black Hat — Menlo, Sweet Security, and Acalvio All Shipped Rogue-Agent Controls This Week
SecurityWeek