Can Ultra Security's $10M Seed Round Keep AI Agents From Going Rogue?

Ultra Security, a Detroit-based agentic security company, has emerged from stealth with a $10 million seed round to secure AI agents while they work. Ludlow Ventures, Hidden Capital, Entrée Capital, Detroit Venture Partners, firedrop and FDVC participated, along with angel investors from the security and SaaS communities, according to the company's launch announcement.

Ultra was founded in 2025 by Chase Lee, who previously founded compliance startup Trustpage, which was acquired by Vanta, alongside a team of security, compliance and engineering veterans.

What Ultra Security Does

Ultra sits between AI agents and the MCP servers and connectors they call. Every tool call, data read and prompt request passes through the platform, giving security teams a live view of what agents are doing and the ability to stop them mid-action.

The product is organized around four functions. Guardrails let companies write plain-language policies that block credential access, redact personal or health data, rate-limit actions or stop data from moving between connectors. Governance controls which agents, tools and connectors each person or team can use. Observability logs and traces every agent-to-tool interaction, with AI-native anomaly detection and alerts. Connectivity acts as a vendor- and model-agnostic bridge between agents and the tools they rely on.

The platform runs locally on the customer's own machine by default, with a remote-hosted option. Ultra says it holds SOC 2 Type II, ISO 27001 and ISO 42001 certifications and has completed a FedRAMP 20x Class B audit as it works toward FedRAMP certification.

"Agents aren't predictable. Security should be," the company said.

Why It Matters Now

Can a seed-stage startup keep agents in line? The case for Ultra rests on where it sits. Traditional security tools watch networks, endpoints and identities, but an AI agent with legitimate credentials can still pull the wrong file, leak customer data into another app or take an action nobody approved. Policing the tool call itself, at the moment an agent tries to act, is one of the few places to catch that behavior before it does damage.

That problem is growing fast. Companies are wiring agents into email, CRMs, code repositories and internal databases through MCP connectors, and security teams are often finding out after the fact. As FinTech Global put it, AI agents are outrunning existing defenses.

Ultra is leaning hard into its positioning, describing itself as "a security company, not an AI infrastructure or orchestration company that added a security tab." The compliance certifications, unusual for a company just out of stealth, are aimed squarely at regulated buyers in finance, healthcare and government, where an agent mishandling patient or account data is a reportable incident rather than a bug.

Detroit is also part of the story. Ludlow Ventures and Detroit Venture Partners have spent years building the city's startup scene, and a security company with federal ambitions gives that ecosystem a foothold in one of the most competitive corners of AI.

The company has not disclosed customer numbers, saying only that it worked with customers and design partners while in stealth. The next test is whether security leaders treat agent runtime protection as its own budget line, and whether Ultra can claim that budget before larger platforms bundle the feature in.

Source

EIN Presswire — "Ultra Security Emerges from Stealth with $10M Seed Round to Secure Autonomous Runtimes"