AI Agent Security Startups Are Piling In, and Reco Just Got Paid

Reco just closed a $55 million round, and the timing says a lot. The raise builds on a $30 million Series B from February, and AT&T, which is also a customer, invested through its venture arm alongside Forestay and Quadrille Capital. As TechCrunch reported, investors clearly want a piece of whoever can secure AI agents.

If the term still feels fuzzy, this explainer on what AI agents actually are helps. They’re also showing up in odd corners now, from finance ops to retail, where fashion agents are a good example.

The numbers behind the raise

CEO Ofer Klein says the valuation has more than doubled since February and sits in the “high hundreds of millions.” Annual recurring revenue is in the double-digit millions, and he expects it to triple this year. Reco has more than 100 customers, and financial services make up about 40% of the business.

Banks and insurers tend to feel this problem first. They have the most to lose and the strictest auditors.

Everyone sells the same story

Read a few vendor pages and you’ll notice the overlap. Discovery, governance, continuous monitoring, runtime security and MCP vetting show up in nearly every pitch, even though the products differ. So the useful question isn’t who says “AI agent governance.” It’s who can actually see the agents.

AI agent security starts with a headcount

Reco’s own examples make the point. At one Fortune 100 client, it found 21,000 agents nobody knew about. At a financial services firm, it found an agent built by an ex-employee that still had access to Salesforce.

That second story is not a clever hack. It’s an offboarding failure. Somebody left, and the thing they built stayed behind with the keys.

It also shows why agents need different handling than a chatbot or a plug-in. A tool waits for you to click. An agent goes off and acts. If you want the longer version, here’s a breakdown of the difference between AI tools and AI agents.

Who owns the agent’s identity?

Every agent runs on credentials, and those credentials don’t belong to a person. Most companies still manage access as if they do. That’s why many security teams now argue for treating AI agent identity separately from human identity and access control.

Reco frames it similarly. In its launch post, it says the goal is to understand what an agent can do, not only what it already did. It maps each agent across identity, permissions, connectivity, and activity. You can read that in Reco’s own announcement. Where you stand decides what you see

Vendors are splitting up the territory. Some vet the tools agents use, some limit the data agents can reach, and some, like CrowdStrike, build detection and response on the devices agents run on. Others hunt unapproved AI use.

Reco’s bet is SaaS. Its context graph links agents, apps, accounts, and permissions, with integrations across more than 280 applications.

My take: an agent living inside a SaaS platform can be nearly invisible to an endpoint tool, and vice versa. Add the sheer variety of agentic AI frameworks teams build on, and no single vendor sees everything.

What to watch next

Three things I’d keep an eye on:

• Discovery will get cheap. Finding agents is the easy part to copy. Blocking or revoking them in real time is harder.
• Platforms will bundle. One report says Reco has a ServiceNow partnership that could give it distribution inside large enterprises. That could be a strength or an acquisition signal.
• Consolidation looks likely. This many similar startups probably can’t all survive.

The takeaway

The $55M says less about Reco than about the problem. Companies are deploying agents faster than they can track them. Whoever answers “what agents do we have?” first gets the chance to answer “what should they be allowed to do?”

 Related:  FancyAI Agents Review: 7 Things You Should Know Before Using Them