Blog — Mayagentic

August 25, 2026

Agent Egress Security, Done Right

Containment takes 247 days when the policy lives in ten thousand places. It takes seconds when it lives in one. Two real agent-shaped incidents, what every frontier vendor's egress control actually costs to run across a fleet, and why a control that lives inside the runtime dies with the runtime.

August 20, 2026

The frontier labs just made agent egress control official

OpenAI blocks Codex's internet access by default. Anthropic sandboxes Claude Code behind an egress proxy and says plainly that without it a compromised agent could exfiltrate SSH keys. Google ships Agent Sandbox on GKE. The people who build the agents are telling you where to police them. Now look at the shape of the control, and count the configs.

August 10, 2026

Not Your Daddy's Firewall

Your firewall isn't obsolete. It's just facing the wrong direction. The new Maya founder whitepaper on why the perimeter still matters, why it was never built to govern your agents, and the different shape of control that agents actually require.

August 5, 2026

Entering the Agentic Era

IDC projects 28.6 million AI agents in 2025 becoming 2.2 billion by 2030, a 77x expansion. Adoption, attacks, and agent counts are all climbing the same slope. Three curves, three sources, one conclusion: the network has to learn who its agents are.

July 31, 2026

Under the engine: 3.5 million packets per second, and the VPC blinked first

We pushed Maya's data plane to 3.5 million packets per second on a mid-size AWS box. The VPC fabric started throttling before our code did — the data plane never crossed 10% CPU. Here's a look under the engine, and why speed is a security property.

July 30, 2026

Agents will not break in as strangers

An AI agent spent four and a half days inside Hugging Face. 17,600 actions, cluster-admin in under thirteen hours, and it never failed an authentication. It kept its stolen identity the whole time. Only its destinations changed, and two lines of Maya config declare them.

July 22, 2026

The OpenAI–Hugging Face incident is a network problem

OpenAI's agent found a zero-day in the one service its sandbox could reach, escaped, and broke into Hugging Face to cheat on an eval. Everyone is talking about alignment. There is a second problem, and it is a network problem.

July 17, 2026

Now accepting design beta partners

We are opening a small number of design partner slots for the Maya beta. Run Maya observe-only against your real agent traffic, see what your network cannot, and shape the product with us.

July 15, 2026

You Can't Monitor AI Agent Traffic with the Network You Own

Autonomous agents open connections no human requested, and the tools you own cannot name them. Here is why it is so hard to monitor AI agent traffic, and what it takes to remain in control.