Stack layer / Threat pattern
BekchiAI Releases 2,057 Verifier-Checkable Agent Tasks Where Gold Answers Are Computed, Not Written
arXiv 2608.26867
Stack layer / Threat pattern
Pentest Harness reaches 306 stars as a self-hosted, bring-your-own-key agent harness for authorized engagements
GitHub
Policy dependency / Threat pattern
94 to 100% of Security Issue Arrivals Land in Queues Already At or Over Capacity, Making Remediation a Flow Problem Not a Ranking One
arXiv 2608.28509
Stack layer / Contrast
Simon Willison reverse-engineers ChatGPT Work and calls out OpenAI for hiding the system prompts and tool descriptions
simonwillison.net
Stack layer / Contrast
Tencent Open-Sourced a 770B Model and Put It Free Inside Its Own Productivity Suite for Two Weeks
Tencent (corroborated by Hacker News front page, 328 points, 2026-08-29)
Stack layer / Threat pattern
OpenClaw 2.0 Ships as an Open-Source Agent Platform That Installs on Your Existing Subscription Instead of Selling One
OpenClaw
Stack layer / Threat pattern
kru is a local-first MCP credential manager that lets agents use secrets without seeing plaintext
GitHub
Stack layer / Threat pattern
An agent's own instruction arbitration swings 47 points between deployment windows, so the trust boundary has to live outside the model
arXiv 2608.28502