doxx.net Raises $38 Million to Prevent AI Agent

doxx.net raised $38 million in a Series A funding round led by Andreessen Horowitz, with participation from Animo Ventures and Focal.vc. The funding coincided with the announcement of doxx.net's Agentic Defined Networking (ADN) platform, which is currently in open beta. The ADN platform provides an environment for agents with defined connectivity and built-in threat protection to prevent them from reaching malicious destinations while browsing and acting on a user’s behalf.
Barrett Lyon, founder and CEO of doxx.net, stated that the industry markets privacy as a feature while using rented infrastructure without ownership or verification. He asserts that doxx.net built a platform where privacy is inherent in the architecture, allowing users and agents to set their own privacy terms rather than relying on third parties.
The ADN platform functions as a parallel, private networking system enabling human users and AI agents to connect, communicate, and browse without central intermediaries. It offers serverless P2P communication with end-to-end messaging and file transfers, threat protection against malicious destinations, agent integration, and privacy features.
Joel De La Garza of Andreessen Horowitz described doxx.net as building a privacy stack from first principles, aiming to deliver security and communications capabilities based on private architecture rather than marketing promises. The platform supports a parallel internet ecosystem with custom TLDs and its own certificate authority, allowing customers to run on infrastructure owned by doxx.net with their own IP space and ASN.

* doxx.net raised $38 million in a Series A funding round.
* The funding was led by Andreessen Horowitz, with participation from Animo Ventures and Focal.vc.
* doxx.net was founded in 2025 by Barrett Lyon.
* The doxx.net Agentic Defined Networking (ADN) platform was announced concurrently with the funding.
* ADN provides agents with defined connectivity and built-in threat protection to prevent reaching known malicious destinations.
* Barrett Lyon stated that privacy should come from the architecture rather than policy documents.
* ADN is a parallel, private networking platform allowing human users and AI agents to connect without central intermediary servers.
* ADN features serverless P2P communication, end-to-end messaging/file transfers, threat protection, agent integration, and anti-tracking features.
* The platform supports 196 custom TLDs and its own certificate authority for a parallel internet ecosystem.
* doxx.net claims to have blocked more than 38 million threats in closed beta since December 2025.
* Customers can run on infrastructure owned by doxx.net with their own IP space, ASN, DNS root, and local AI models.

The narrative positions a shift from privacy as a policy-based feature to privacy as an architectural necessity, spearheaded by the creation of a parallel networking layer. This structure attempts to delegitimize the conventional model where infrastructure providers control privacy terms by framing them as inherently compromised or rented. The concept of "privacy by architecture" suggests that security and privacy should be foundational constraints rather than optional overlays.
The mechanism described—a decentralized, P2P system for agents and humans operating outside centralized surveillance—presents a fundamental tension between the desire for ubiquitous internet access and the need for secure, private interaction. The claim of blocking 38 million threats in testing establishes an empirical metric for the effectiveness of this architectural approach, but it raises questions about the scope and permanence of these protections across diverse network conditions.
The implications center on sovereignty: when infrastructure itself is owned by the users or agents via a self-defined network architecture, control over information flow shifts from intermediaries to the participants. This challenges the established power structures that rely on centralized oversight for maintaining digital order. The core question becomes whether distributed, owner-controlled networking can scale effectively against increasingly complex adversarial tactics, and who ultimately bears the responsibility when emergent threats bypass these defined boundaries?
Bridge Questions: If privacy is truly defined by architecture, what metrics should govern the interoperability between sovereign networks? How does this P2P model address censorship imposed by state actors on non-public routing? What are the unforeseen systemic costs of creating a parallel internet ecosystem governed by custom TLDs and decentralized certificate authorities?