From the Research Desk

Written by the people
who break things first.

Notes on defending networks, attacking them to prove what is real, and keeping AI agents honest — from the team that builds Maximus.

Illustration representing air-gapped AI security for mission-critical infrastructure
Sovereign AI

Why Mission-Critical Infrastructure Needs Air-Gapped AI Security

The places with the most to lose — defense networks, power grids, factories — are the ones that can't use cloud AI. Here is how to fix that.

July 28, 2026 · 8 min read
Illustration of an agentic security operations center
AutoSecOps

The Agentic SOC: When Security Operations Run Themselves

What changes when your security tool does the work instead of handing you a to-do list.

July 14, 2026 · 10 min read
Illustration of AI agent runtime governance
AIDR

Your AI Agents Have Shell Access. Who Is Watching Them?

They have your passwords and your servers. Nobody is checking what they do with them.

June 30, 2026 · 9 min read
Illustration of autonomous offensive security validation
BreachOps

Proof Over Alerts: The Case for Continuous Offensive Validation

Your scanner found 4,000 problems. An attacker needs one. Only an attack tells you which.

June 16, 2026 · 7 min read
Abstract illustration representing the Maximus cyber-reasoning model
Maximus

Benchmarking Cyber Reasoning: Inside Maximus, Cybench 91% and CyberGym 0.78

What these tests actually measure, and how a model with no internet beats ones that have it.

June 2, 2026 · 11 min read