Illustration for: doxx.net Raises $38M To Rein In AI Agents

doxx.net Raises $38M To Rein In AI Agents

doxx.net raised a $38 million Series A led by Andreessen Horowitz to build tighter authorization and oversight controls for autonomous AI agents acting on users' behalf.

By the Numbers

$38M
Series A
Andreessen Horowitz
Lead investor
ANIMO Ventures, Focal.vc
Other backers
AI agent oversight
Focus
TC
Early-stage VC & angel · Founder, New York Venture Partners · Value Add Pulse Funding Desk
1 min read
ShareXLinkedInEmail

THE RUNDOWN

1

As AI agents gain the ability to make purchases and manage accounts autonomously, tools that constrain and audit what those agents can actually do are becoming their own funding category.

2

Andreessen Horowitz leading the round adds to the firm's broad bet across agent infrastructure, from payments rails at companies like Jeeves to doxx.net's permissioning layer.

3

The framing -- putting AI agents on a shorter leash -- speaks directly to enterprise buyers still nervous about giving autonomous agents real financial or account access without guardrails.

4

Watch whether doxx.net's approach becomes a standard layer every agent platform licenses, or whether larger players like OpenAI and Anthropic build equivalent permissioning natively and squeeze out standalone vendors.

TC

The VC Read · Trace's Take

Trace Cohen

Every agent-infrastructure startup pitching guardrails lives or dies on one question: does a major model lab ship this natively in six months? I'd diligence whether doxx.net's permissioning layer is model-agnostic enough to survive OpenAI or Anthropic building the same thing into their own agent SDKs.

Analysis

doxx.net has raised a $38 million Series A led by Andreessen Horowitz, with participation from ANIMO Ventures and Focal.vc, Refresh Miami reported, to build infrastructure that authorizes and constrains what autonomous AI agents are allowed to do on a user's behalf.

The round lands as AI agents move from chat-based assistants into systems that can make purchases, manage accounts and execute multi-step tasks without direct human approval at each step -- the same shift driving rounds for companies like Instinct and Beltic, both building adjacent pieces of agent infrastructure this year. doxx.net's pitch is narrower: rather than building the agent itself, it builds the permissioning and oversight layer that sits between an agent and the actions it's allowed to take. Miami-based ANIMO Ventures' participation continues a pattern of local funds co-investing alongside major Silicon Valley firms on South Florida-adjacent deals, though doxx.net itself hasn't disclosed where it's headquartered.

“That's a crowded and fast-moving lane, one that overlaps with the broader rush of 2026 agent-infrastructure financing chronicled across this issue.”

That's a crowded and fast-moving lane, one that overlaps with the broader rush of 2026 agent-infrastructure financing chronicled across this issue. Beltic is building infrastructure to verify and authorize autonomous agents with a similar pitch, and the major model labs -- OpenAI and Anthropic chief among them -- have strong incentive to build equivalent permissioning natively into their own agent SDKs rather than cede the layer to a third party. doxx.net's survival depends on becoming genuinely model-agnostic enough that enterprises adopt it across multiple agent platforms rather than treating it as a stopgap a lab's own tooling eventually replaces.

See our AI valuations tracker for how agent-infrastructure rounds like this one compare to the broader 2026 AI financing wave. doxx.net hasn't disclosed total funding, headcount or customer counts, and Refresh Miami's report doesn't name any launch customers or say which agent platforms doxx.net's permissioning layer currently integrates with -- the detail that would actually show whether the model-agnostic pitch holds up in practice.

ShareXLinkedInEmail

Key Sources

2 sources

THE WIRE in your inbox— Tech, startup & VC news with Trace's take, a few times a week. Free to subscribe, no spam.