What Zenity's $125M Series C Actually Tells Us About AI Agent Governance

Zenity closed a $125 million Series C this week. Norwest Venture Partners led it, with SoftBank Vision Fund 2, Qumra Capital, Hitachi Ventures, and LG Technology Ventures joining, alongside existing backers Vertex Ventures, Third Point Ventures, DTCP, and Intel Capital. That brings the company's total funding to $185 million. Back in April, Gartner named Zenity the company to beat in AI agent governance.

I read the announcement three times before I fully absorbed what it meant, and not because of the number itself. Founders get numb to big funding headlines eventually. What stopped me was thinking about who actually writes checks that size, and what they had to believe was true before they did.

Norwest didn't fund this because AI agents might someday need governance. Growth-stage investors don't operate on someday. They operate on evidence sitting in front of them right now, usually a pipeline of customers already paying for something and asking for more of it. If a fund with Norwest's diligence bandwidth is putting nine figures behind this, the underlying demand is not speculative anymore. It's already showing up in enterprise budgets.

A 2026 Economist Enterprise survey backs this up with numbers that are honestly a little alarming. Among more than 800 senior technical executives at companies with over $500 million in revenue, nine in ten said they're deploying AI agents faster than their security teams can secure them. Nearly all of them had already had an agent-related incident disrupt their organization in some way. Only three in ten had a rollback process they actually trusted if something went wrong.

Read those three numbers together and you get a fairly uncomfortable picture. Most enterprises are moving fast, most have already been burned once, and most still don't have a reliable way to undo the damage if it happens again. Zenity's raise is the market putting a price on that gap.

Two different bets on what "governance" means

Here's where it gets more interesting than the funding figure, at least to me. There are two fairly distinct ways companies in this space have chosen to solve the problem, and they're not really competing answers to the same question. They're answers to two different questions that happen to sound similar.

One approach tries to understand what an agent is doing and why, distinguishing normal behavior from something manipulated, compromised, or drifting outside its intended purpose. This is fundamentally a judgment call, made well by a system that's read a lot of context. Zenity's research team has done strong work here. Their documentation of attack patterns like prompt-injection-driven data exfiltration through connected services is genuinely useful reading if you haven't seen it.

The other approach skips the judgment call entirely. Every action gets checked against explicit rules. No model interprets intent. No probability score decides anything. The verdict is allow, hold for a human, or block, and the same input produces the same output every single time, which sounds almost too simple to matter until you try explaining a probabilistic decision to a bank's internal auditor.

I'd push back gently on treating these as "better" and "worse." They're suited to different jobs. If you're hunting for a novel attack pattern nobody has documented yet, you want a system that can reason about intent and context. If you're a compliance officer who has to sign a document saying your AI systems didn't do anything unauthorized last quarter, you want something that produced the same answer regardless of who was asking or how the input was phrased. Those are different problems, even when the marketing copy for both makes them sound identical.

Why the distinction gets sharper as the category grows

As AI agent governance moves from an emerging worry into a funded, Gartner-tracked category, buyers are going to sort themselves by which of these two questions they're actually trying to answer. A security operations team wants the richest possible signal on anomalous behavior, full stop. A compliance officer preparing for an EU AI Act Article 12 audit wants something narrower: not a rich signal, a defensible one.

I've seen this from the regulator's side of the table too, which is probably why I notice it. Years ago I sat on Pakistan's SECP Fintech Working Group, part of the small group that helped draft the country's national fintech regulatory framework. Before that, I'd built a real estate co-investment platform that the same regulator flagged publicly within days of launch. So I've had the specific, unpleasant experience of sitting across from a regulator who wants a straight answer to a straight question, and I've also had the experience of being the person drafting what counts as an acceptable answer. Those two chairs taught me something that doesn't show up in most product pitches: a regulator does not want your best guess, however well-reasoned. They want proof, in a form they can independently verify without trusting your interpretation of it.

That is the specific, narrow problem I built Gateplex to solve. Every agent action gets intercepted before it executes and evaluated deterministically, with no model anywhere in the decision path. Every decision lands in a tamper-evident, hash-chained audit trail, and it can be exported as signed compliance evidence formatted for the people who actually have to stand behind it in a room: legal teams, auditors, regulators. Not a dashboard built for the engineer who wrote the agent.

The category is real now. The specific need still isn't fully solved

Zenity's raise is good news, honestly, including for us. It confirms the problem is real, funded, and taken seriously by people whose job is skepticism. It doesn't mean the problem is solved for everyone deploying agents into regulated workflows.

As more enterprises push agents into financial transactions, healthcare records, legal processes, the question that keeps resurfacing isn't "did something anomalous happen." It's narrower than that, and harder to answer well: when an agent takes an action, can you actually prove what it did and why it was allowed to do it, to someone whose entire job is not taking your word for it.

The market just decided that question is worth $125 million to work on. I think it's worth working on precisely, not just at scale.

Umair Sheikh is the founder and CEO of Gateplex, a real-time governance firewall for autonomous AI agents. Learn more at gateplex.ai.