Policy dependency / Stack layer
Unsloth 0.1.805-beta doubles Qwen3.8-Flash-Next and GLM-5.3-Flash inference with MTP on by default
GitHub
Stack layer / Threat pattern
CIPR: how you phrase a request changes whether a poisoned repo compromises your coding agent
arXiv
Stack layer / Threat pattern
OpenAI Says Astra Is Its First Model to Cross the Critical Cyber Threshold, and It Is Shipping It Behind Gates Anyway
OpenAI / Axios / CSO Online
Stack layer / Threat pattern
Browzer Took Product Hunt's Top Spot Selling Self-Healing Docs Against Mintlify and GitBook
Product Hunt
Stack layer / Threat pattern
Anthropic restarted external cyber evals with a classifier that kills a tool call when a model tries to escape the sandbox
Anthropic
Policy dependency / Stack layer
Vercel's AI SDK harness layer now runs nine coding agents behind one API, including its own fx
Vercel Changelog
Policy dependency / Stack layer
WebWorld Uses the Browser as the Judge a VLM Cannot Fool, Beating Its Own Base Model by 14.9 Points
arXiv 2608.30530
Policy dependency / Stack layer
Jamf enforces per-user Bedrock spend caps in minutes using IAM policy rewrites, for under $10 a month
AWS Machine Learning Blog