Microsoft disclosed a maximum-severity Entra ID flaw: unauthenticated RCE via unsafe deserialization, CVSS 10.0. Microsoft fixed it server-side — here's what defenders should still do.
An autonomous AI agent chained two bugs in Hugging Face's dataset-processing pipeline, stole cloud credentials and moved laterally through production. What it means for your ML pipelines and agents.
CISA added CVE-2026-72529 (missing auth, CVSS 9.3) and CVE-2026-72530 (code injection, CVSS 9.0) to KEV. The Head Mare group is swapping the client installer for a backdoor. Breakdown and defense.
274 Zimbra servers already compromised via CVE-2026-73570 — an SNMP-component command injection giving unauthenticated RCE. The exact preconditions, why patching isn't enough, and what to do.
CloudSEK's exposed-server analysis shows a Russian-speaking Aurora affiliate using the agentic assistant Cursor to plan intrusions and Active Directory escalation. Two defender lessons and what to do about them.
On 27-28 Aug 2026 PaperCut confirmed active exploitation of a zero-day in NG/MF (all versions affected) and shipped an emergency patch. We break down why print servers are prime targets, the IOCs, and a…
Security skills for AI agents are now a whole category — and against prompt injection they barely help. Here's why a skill is just more prose on the layer injection attacks, and why a…
Airlock is an open-source runtime firewall for AI coding agents — Claude Code, Cursor, MCP. Gate every tool call, stop tool poisoning, keep a tamper-evident log.