Neo’s $100M Bet: Enterprise AI Agents Need Their Own Security Stack

One hundred million dollars buys a lot of things. A small island. A decent private jet. Or, if you’re Neo, a war chest to solve the problem that keeps CISOs awake at 3 a.m.: how do you secure software that writes its own code, spins up its own infrastructure, and negotiates its own API contracts — all without a human in the loop?

The Agentic AI Security Gap Is Real

Let’s be honest about where we are. Enterprises aren’t just “experimenting” with AI agents anymore. They’re deploying them in production — handling refunds, provisioning cloud resources, triaging security alerts, even drafting legal contracts. These aren’t chatbots. They’re autonomous actors with API keys, database credentials, and the ability to chain actions across systems.

Traditional security tools were built for deterministic code. Static analysis, SAST, DAST, WAFs — they assume the logic doesn’t change between deployments. An agent that rewrites its own prompt chain at runtime? That breaks the model. Neo’s funding round, led by Lightspeed Venture Partners with participation from Menlo Ventures and others, signals that investors finally grasp the magnitude of this mismatch.

Why Existing Tools Fall Short

I’ve watched security teams try to retrofit existing stacks for agentic workloads. It’s painful. You end up with:

• Runtime application self-protection (RASP) tools that can’t see inside the agent’s reasoning loop
• API gateways that treat every agent call as a generic “user” request, losing context about which subtask triggered it
• SIEM correlation rules that drown in noise because each agent action looks like a separate event
• Identity systems that can’t express “this agent acting on behalf of user X for purpose Y within scope Z”

The result? Security teams either block agents entirely (killing ROI) or YOLO them into production with crossed fingers. Neither scales.

What Neo Is Actually Building

Neo’s pitch isn’t “we’ll scan your agents.” It’s “we’ll give you a control plane for agent behavior.” Think: policy-as-code for what agents can do, not just what they are. Continuous verification that the agent’s runtime behavior matches its declared intent. Cryptographic attestation chains linking every tool call back to the original human authorization.

If they pull this off, it’s the difference between “we hope the agent doesn’t delete the production database” and “the agent mathematically cannot delete the production database because its capability token lacks that scope.” That’s a fundamental shift from detection to enforcement.

The Enterprise Adoption Blockers

Here’s what nobody says at conferences: the biggest barrier to agentic AI in Fortune 500 companies isn’t model capability. It’s liability. Legal and compliance teams need to answer: “When the agent hallucinates a vendor payment and sends $2M to the wrong account, who’s responsible? How do we prove it wasn’t an insider? How do we demonstrate due diligence to regulators?”

Neo’s approach — if it works — turns those questions from existential risks into auditable logs. That’s the real unlock. Not better models. Auditable agency.

Where This Goes Wrong

Skepticism is warranted. The agentic security space is littered with startups that built beautiful dashboards for problems that don’t exist in production. Neo faces three execution risks:

• Integration depth: They need hooks into every major agent framework (LangChain, AutoGen, crewAI, custom runtimes) and every enterprise identity provider (Okta, Entra, Ping, custom LDAP). Missing one kills deals.
• Performance tax: If their verification layer adds 200ms latency per agent hop, high-frequency agents (trading, fraud detection, real-time bidding) become unusable.
• Policy expressiveness: Enterprises have complex, nested, exception-ridden policies. “Agents can read customer data except PII unless it’s a GDPR request handled by the legal agent between 9-5 on weekdays.” If Neo’s policy language can’t express that natively, security teams will write workarounds that defeat the purpose.

The Market Signal

$100M Series B (bringing total funding to $150M) in this climate means sophisticated investors see a category creator, not a feature company. They’re betting that “agentic AI security” becomes its own budget line item — like SIEM, EDR, or CSPM — within 24 months. I think they’re right.

Every enterprise I talk to has the same roadmap: pilot agents in Q1, hit a security wall in Q2, scramble for solutions in Q3. Neo’s timing aligns with that wall. The question isn’t whether this market exists. It’s whether Neo can move from “impressive demo” to “deployed in three clouds with zero-severity incidents” faster than the incumbents (Palo Alto, CrowdStrike, SentinelOne) bolt on agent modules to their existing platforms.

What I’m Watching

Next six months will tell. Watch for:

• Production case studies with named enterprises (not “a major financial institution”)
• Open-sourced policy language or SDK — community adoption predicts platform stickiness
• Integration partnerships with the agent frameworks themselves, not just cloud providers
• Hiring: are they adding runtime systems engineers or just sales reps?

Agentic AI isn’t slowing down. The enterprises that figure out safe deployment first will compound advantages — faster iteration, lower operational costs, new product categories. Neo just got the capital to be the guardrail that makes that speed possible. Now they have to build it.

Source: original report