Stack layer / Threat pattern
Salesforce frames its agent stack as an 'Enterprise AI Harness' with a separate AI Control Plane
Salesforce
Stack layer / Contrast
humans& Released Persimmon, a 550B User Simulator That Fools LLM Judges About 20% of the Time in Group Chats
humans&
Stack layer / Threat pattern
Anthropic Walked Away From a ~$6B Decart Deal Post-LOI, and 20VC's Read Is That the Tech Did Not Generalize
Bloomberg (corroborated by The Next Web, Calcalist, Ynet and PYMNTS; the diligence read is from SaaStr/20VC, 2026-09-11)
Stack layer / Threat pattern
MCPHub before 1.0.32 lets an intercepted OAuth authorization code be redeemed for tokens
NVD
Policy dependency / Stack layer
Mullenweg Tells Automattic Staff in Slack He Is Back in Control, Days After the Board Voted Him Out With 50 Minutes' Notice
TechCrunch (corroborated by 404 Media, Search Engine Journal and NewsBytes)
Stack layer / Update thread
Tailscale gates customer model access by tailnet identity and issues no API keys to agents at all
Vercel Blog
Threat pattern / Contrast
Anthropic's September Threat Report: A Stolen Developer Token Became Full Cloud Admin in About Three Hours
Anthropic (corroborated by SiliconANGLE, TechCrunch and The New York Times)
Stack layer / Update thread
100 LLM agents running a simulated town economy for 26 weeks froze the money supply: 0.3% of prices ever changed, and memory deletion made no difference
arXiv