Prolexic Founder Barrett Lyon Raises $38 Million From a16z for doxx.net, a Private Network Built for AI Agents

doxx.net, a Miami-based startup building what it calls "Agentic Defined Networking," has raised a $38 million Series A led by Andreessen Horowitz, with AnimoVentures and other investors participating, as the platform launches into open beta. The company is the latest bet that AI agents browsing, buying and messaging on our behalf will break the internet's existing privacy and security assumptions.

The founder is a repeat infrastructure builder. Barrett Lyon created Prolexic Technologies, one of the first commercial DDoS mitigation networks, which was acquired by Akamai; he went on to found BitGravity (sold to Tata Communications), XDN (sold to Fortinet) and Defense.net (sold to F5), and has worked with law enforcement tracking extortion crews. That résumé matters, because doxx.net's core product is defensive: a private network layer designed for both people and the agents that act for them.

The pitch is that AI has inverted the privacy calculus. Agents gathering context on a user's behalf can leak far more personal data than a human with a browser, and every agent delegated with real authority expands the attack surface of its owner. doxx.net's answer is a parallel network where people and their agents can browse privately, transfer files peer-to-peer without a server in the middle, and join encrypted device-and-agent meshes — with serverless P2P calling and messaging as first-class services.

The company owns its entire stack rather than renting it: its own IP space and ASN, its own DNS root, its own certificate authority, its own local AI models and its own bare-metal servers across six continents. The platform offers 196 custom top-level domains and its own naming system, so the network resolves domains the public internet has never heard of, with no third party able to see them. Anti-fingerprinting masks browser and device identifiers, and traffic moves through onion-style routing so no single point knows both who you are and what you are doing.

There are no usernames, passwords, email addresses or phone numbers: accounts are created via proof-of-work, and the open beta asks for nothing more than a visit to app.doxx.net — or, in a sign of who the company thinks its users are, "have your agent visit config.doxx.net." Native apps ship for iOS, macOS and Android, with Windows and Linux support running on the same global backbone. The company says it has blocked more than 38 million threats since December.

For AI-agent builders, the notable piece is the agent-native networking model. An agent on doxx.net can operate under its owner's network identity with scoped permissions, rather than scraping the open web from a data center IP that sites increasingly block — a friction point that has grown all the way up to major platforms shutting down public APIs and RSS feeds over bot traffic. A network that can distinguish "my principal's agent" from "some scraper" is, in effect, selling an identity layer for the agentic web.

The round lands in a market where agent infrastructure funding has been accelerating, from durable-execution systems to agent sandboxes and agent-native payment rails. Whether doxx.net becomes the privacy fabric for consumer AI agents or remains a power-user niche will depend on whether mainstream agent platforms adopt its protocols — but with a four-time infrastructure founder, a16z's lead, and an open beta that deliberately asks for nothing, the company is betting that privacy-by-architecture becomes a requirement before agents go mainstream.