doxx.net Raises $38 Million to Prevent AI Agent-on-the-Internet Misadventures

A $38 Million Bet on Keeping AI Agents on a Leash

Miami-based cybersecurity startup doxx.net has raised $38 million in a Series A funding round led by Andreessen Horowitz, with participation from Animo Ventures and Focal.vc. The raise was announced alongside the open-beta launch of the company's new Agentic Defined Networking (ADN) platform, built to address a problem the company says is quietly growing as AI agents take on more autonomous tasks: agents acting on a user's behalf often cannot reliably tell the difference between a safe destination and a malicious one, even though they are operating with that user's accounts and authority.

Joel De La Garza, a partner at Andreessen Horowitz, is joining doxx.net's board as part of the deal. The announcement, dated October 1, 2026, positions doxx.net at the intersection of two trends security teams are tracking closely: the explosion of agentic AI deployments and the resurgence of privacy-first network infrastructure.

Details

What doxx.net's ADN Platform Does

The Problem It Targets

As AI agents increasingly browse the web, call APIs, communicate, and complete tasks on a user's behalf, they do so with that user's accounts and permissions. doxx.net argues that most of today's internet infrastructure was never designed with this scenario in mind: an agent has no reliable, architecture-level way to distinguish a legitimate destination from a phishing page, a malware host, or a spoofed service before it acts. A single bad judgment call by an agent can translate directly into account compromise, data exfiltration, or financial loss for the human it's acting for.

How ADN Is Built

Rather than layering policy or filtering rules on top of the existing internet, doxx.net says it built ADN on infrastructure the company owns end to end, including:

• Its own IP space and ASN
• Its own DNS root
• Its own certificate authority
• Its own local AI models
• Its own bare-metal servers
• A serverless, peer-to-peer communications layer for calls, video chat, messaging, and file transfer that routes directly between devices instead of through a central relay
• 196 custom top-level domains as part of a parallel addressing system
• Account creation via proof-of-work rather than usernames and passwords

The pitch is that agents (and human users) operating inside ADN get a network environment with defined, vetted connectivity and built-in DNS-level threat protection, rather than relying on the open internet's existing trust assumptions. During a closed beta that started in December 2025, doxx.net says the platform blocked more than 38 million potential threats — a company-reported figure describing blocked destinations and attempts rather than confirmed, investigated attacks.

Founder's Framing

Founder and CEO Barrett Lyon — whose prior companies have been acquired by Akamai, Tata Communications, Fortinet, and F5 — framed the approach as a rejection of privacy-as-a-feature marketing: "The industry sells privacy as a feature while routing your traffic through rented infrastructure they don't own and can't verify. We physically built something different: a platform where privacy comes from the architecture itself, not a policy document."

Impact Assessment

Recommendations

For Security and IT Leaders

• Inventory which AI agents in your environment currently operate with live user credentials or OAuth-granted access, and map what those agents can reach unsupervised.
• Treat any agent acting "on behalf of" a user as a privileged identity requiring the same monitoring, least-privilege scoping, and egress controls as a service account.
• Evaluate emerging agent-security tooling (including ADN) against existing DNS filtering, secure web gateway, and zero-trust network access investments before adding a new infrastructure layer.

For Organizations Piloting Agentic AI

• Require logging and alerting on every destination an autonomous agent reaches, not just the actions it completes.
• Avoid granting agents standing credentials where short-lived, scoped tokens are feasible instead.
• Pressure-test vendor claims (like beta threat-blocking counts) against your own telemetry before relying on them for risk decisions.

For Individual Users

• Be cautious about granting AI agents broad account access (email, banking, shopping) until the agent's guardrails and audit trail are well understood.
• Review any permissions already granted to AI browser extensions or agent tools and revoke ones that aren't actively in use.

Key Takeaways

1. doxx.net raised a $38 million Series A led by Andreessen Horowitz, with Animo Ventures and Focal.vc also participating.
2. The round funds the open-beta launch of ADN (Agentic Defined Networking), built to stop AI agents from reaching unsafe destinations while acting under a user's authority.
3. ADN runs on infrastructure doxx.net says it owns outright — its own ASN, DNS root, certificate authority, and bare-metal servers — rather than on top of existing internet trust models.
4. A closed beta starting December 2025 reportedly blocked more than 38 million threats, though that figure is self-reported and unaudited.
5. The raise reflects growing investor and market interest in treating agentic AI access as a distinct security problem, not just an extension of existing endpoint or network defenses.
6. Organizations deploying AI agents with live user credentials should independently verify any vendor's protective claims and maintain their own visibility into what those agents can reach.

Sources

• SecurityWeek — doxx.net Raises $38 Million to Prevent AI Agent-on-the-Internet Misadventures
• BusinessWire — doxx.net Raises $38 Million, Led by Andreessen Horowitz, to Launch Agentic Defined Networking
• Refresh Miami — doxx.net raises $38M to put AI agents on a shorter leash