One misconfigured DNS resolver let an OpenAI research model reach the internet, and OpenAI paused frontier training for 11 days. What it built next, and four controls enterprises running AI agents should copy: na2.hubs.ly/H08FGzX0

PHP2Shell: patch first. Segmentation won’t block the exploit on a public site, but it can keep an attacker on that one server and block the callback. What Zentera can and can’t do: hubs.ly/Q04Cb4_10 #PHP2Shell #ZeroTrust

NVIDIA has a blueprint for securing enterprise AI agents. Here's how to put it into practice today. hubs.la/Q04zZcLf0

Every AI agent security demo shows you what the product detects. Almost none show you what a compromised agent can still reach when detection misses. That's the question we'd ask first. hubs.la/Q04yYzSk0

Engineers need broad file access. Their AI agents don't. Standard file permissions can't tell the two apart. hubs.la/Q04yBKHZ0 #pressrelease #cybersecuritynews #AIagentSecurity

79% expect to claw back AI agents within 18 months. Fewer than half are confident they can do it selectively. Why the "kill switch" most enterprises have doesn't reach the agents they're worried about: hubs.la/Q04y3JjG0

Only 43% of security leaders are very confident they could prove what an AI agent was authorized to do. Most of the rest aren't sure. Download the infographic: hubs.la/Q04xHbxc0

84% of security leaders say AI agents cross boundaries more easily than their own employees do. Nobody built a security model for that. Download the infographic: hubs.la/Q04xG-f80

87% of security leaders say authorization is the missing layer in agentic AI security. New research on how enterprises are actually governing AI agents: hubs.la/Q04xCMJ90

An AI coding agent deleted a production database and every backup in 9 seconds. No attacker involved. It was just doing its job. That's the Lethal Trifecta: untrusted input, sensitive access, and the ability to act, all in one agent session. hubs.ly/Q04stw990 #AgenticAI #AISecurity #ZeroTrust #LethalTrifecta

Three weeks ago this was one AI lab's bad week. We've updated our piece: it's now four labs, same failure. hubs.la/Q04sqV4B0 #AIAgents #AISecurity #EnterpriseAI #ZeroTrust

The OpenAI model that broke into Hugging Face didn't get smarter when its guardrails came off. The environment around it just stopped holding. That's the whole incident. hubs.la/Q04rvND00 #AIAgents #AISecurity #ZeroTrust

The Hugging Face Breach: 17,000 Actions, One Unprotected Blast Radius youtu.be/cfH-7RNW7Ek?si=bpzI… via @YouTube

Prompt injection. Shadow agents. Uncontrolled connectivity. The new attack vectors AI agents introduce, and the 6-step framework to close the gap without slowing adoption: hubs.la/Q04pyCyX0 #AIAgents #ZeroTrust #CyberSecurity

We will be exhibiting at DAC 2026: The Chips to Systems Conference on Jul 26 - 29, 2026. Like or comment if you'll be there! #DAC2026 hubs.la/Q04pphBt0

An OT assistant in March. An OT agent by May. No procurement event, no architecture review, just an upgrade. CISA, NCSC, and Five Eyes have published 3 joint guidance docs on securing AI agents in critical infrastructure since Dec 2025. Here's what they agree on: hubs.la/Q04pdF890

Server sprawl. Then cloud sprawl. Now agent sprawl, and this one is different: an agent can write another agent. Govern AI agents before they outrun your security team. Live with Zentera's Nathanael Iversen, Jul 28: hubs.la/Q04nHKrX0 #AISecurity #AgenticAI

Zero Trust says verify explicitly. Most enterprise AI agents authenticate once at deployment and run indefinitely under inherited service account credentials. No re-auth, no session expiry, no per-agent audit trail. hubs.la/Q04mTtCt0 #AISecurity #ZeroTrust

Threat Briefing – June 2026 youtu.be/EQXuim8zSno?si=Y2RL… via @YouTube