doxx.net Raises $38M to Build Agentic Defined Networking

An AI agent can browse, call APIs, move files and change settings with a user's authority. The harder question is who controls the network boundary underneath that behavior when the agent can also help configure the network itself.

doxx.net has raised a $38M Series A to answer that question with infrastructure it says it owns end to end. Andreessen Horowitz led the October 1, 2026 round, with Animo Ventures and focal.vc participating. The financing arrived as doxx.net opened its Agentic Defined Networking platform to public beta.

The company is building private networking for people and AI agents across browsing, direct communications, file transfer, device connectivity, naming, routing and threat protection. That puts doxx.net inside a wider shift in AI infrastructure: as agents gain permission to act, the systems surrounding them need to enforce where that authority begins and ends.

What doxx.net announced

doxx.net's $38M Series A is led by a16z, and a16z partner Joel de la Garza is joining the company's board. The company did not disclose a valuation, a cumulative funding total, revenue, customer count or a detailed allocation for the new capital.

Founder and CEO Barrett Lyon brings more than 20 years of internet-infrastructure and security experience to the project. Lyon founded Prolexic, later acquired by Akamai, and subsequently built BitGravity, XDN and Defense.net, which were acquired by Tata Communications, Fortinet and F5. That record does not prove the new architecture, but it explains why the company is starting below the application layer rather than adding another privacy dashboard to rented infrastructure.

Founded in 2025 and based in Miami, doxx.net says its network now operates across 6 continents. The public beta follows a closed-beta period that began in December 2025.

How Agentic Defined Networking works

doxx.net calls its model Agentic Defined Networking, or ADN. People and agents receive network identities and can create private connections, set firewall policy, connect devices, transfer files and communicate directly. Native MCP support gives agents a way to inspect and configure networking through an API under the user's policies.

The company says it owns its IP space, autonomous system number, DNS root, certificate authority, local AI models, bare-metal servers and global backbone. Private names can resolve inside the doxx.net network rather than the public internet, while calls, messages and files can travel directly between peers without a communications server in the middle.

The beta also includes DNS-level threat protection, covert-networking protocols, anti-fingerprinting and onion-routing features. Users can start without supplying an email address or phone number, and the company says accounts are created through proof-of-work. These are architectural choices intended to reduce the identity data a privacy service collects before it begins protecting activity.

Why the agent layer changes the privacy problem

Traditional privacy tools usually assume a person initiates a session, chooses a destination and understands the action being taken. An agent can instead make repeated decisions across websites, APIs, files and accounts while the user is elsewhere. Every new permission expands both usefulness and exposure.

That makes network policy part of the agent stack. The question is no longer limited to whether traffic is encrypted. Operators also need to know which destinations an agent may reach, how its identity is separated from the user's, where policy is enforced, and what infrastructure can observe the relationship between the two.

a16z's investment thesis is that privacy becomes more urgent as AI makes collected data easier to analyze and connect at scale. doxx.net's answer is to give users and agents a parallel network with its own controls instead of asking a conventional network to behave more discreetly.

What the early metrics do and do not show

doxx.net says it blocked more than 38M potential threats during closed beta since December 2025. The figure describes company-reported activity across its protection layer. The company has not published an independent audit methodology that would let outsiders equate those events with 38M prevented attacks or compare efficacy with established security products.

The open beta is therefore the more consequential milestone. It moves the platform from a controlled environment into broader consumer and prosumer use, where reliability, latency, policy clarity and day-to-day usability become visible. Barrett Lyon told Axios that an enterprise offering is planned by the end of 2026, a forward-looking target that would add procurement, administration and compliance demands to an already broad technical surface.

What the $38M round makes possible

Owning network infrastructure creates control, but it also creates cost and operational obligation. A private DNS root, certificate authority, global backbone, bare-metal footprint and multi-platform applications must remain available, secure and understandable while the product asks users to trust a relatively new company with an unusually deep position in their connectivity.

The Series A gives doxx.net room to develop that system beyond its beta without pretending the commercial proof is already complete. The company has not disclosed customer numbers, revenue, retention or enterprise contracts, so the investment is best read as conviction in the architecture and the founder's operating history rather than evidence of a mature market position.

The useful test will happen inside ordinary behavior. If an agent can configure a private route, avoid a malicious destination, transfer a file and communicate with a user without exposing more identity than the task requires, the network becomes an active control layer instead of background plumbing. doxx.net now has $38M and a public beta in which to make that handoff work.