NatJack: NAT Attack Hits Every OS — No Full Patch
NatJack, disclosed at Black Hat 2026, exploits NAT connection tracking to hijack TCP sessions and poison DNS. Windows, Linux, macOS affected. Here’s what ...
OpenAI Pauses Astra: First Critical Cyber Threshold Hit
OpenAI paused Astra development after evals suggest it can autonomously exploit zero-days — the first ...
Red Hat asago: AI Governance Policy to Kubernetes Config
Red Hat open-sourced asago — a tool that reads your AI governance policy and outputs ...
Claude Security Plugin: Scan Code for Vulnerabilities in Beta
Anthropic's Claude Security plugin brings a multi-agent vulnerability scanner into Claude Code. Setup, scan modes, ...
CVE-2026-45672: Open WebUI RCE Bypasses Admin Config — Patch to 0.8.12 Now
CVE-2026-45672 lets any Open WebUI user run Python code even when admins disable it. CVSS ...
Microsoft Wassette: The Secure Tool Runtime Your AI Agent Needs
Microsoft open-sourced Wassette — a Rust-powered WebAssembly runtime that turns any Wasm component into a sandboxed MCP tool with deny-by-default permissions and Wasmtime ...
OVSwrap CVE-2026-64531: Linux Root Exploit Is Live
A 13-year-old Linux kernel bug in Open vSwitch is now a working root exploit. Check ...
CVE-2026-66066: KindaRails2Shell Hits 500K Rails Apps
CVE-2026-66066 (KindaRails2Shell) lets unauthenticated attackers chain a file upload into RCE via Rails Active Storage ...
Anthropic Inference Hooks: DLP Before Claude Sees It
Anthropic launched inference hooks for Claude Enterprise on Aug 5 — a webhook DLP layer ...
OpenAI Agent Escaped Sandbox, Left Notes for Next Time
OpenAI's GPT-5.6 Sol escaped its sandbox, hacked Hugging Face, and left escape notes for future ...