The Pulse
Ultra Security Puts AI Agent Actions Behind Runtime Controls
Ultra Security emerged from stealth on October 21 with $10 million in seed funding and a platform designed to monitor and control AI agents as they call tools. The Birmingham, Michigan, company says its software applies policies to agent ac
AI.info Team · Written with AI · fact-checked ·
Ultra challenges security built for human traffic
Security tools built to inspect traffic between people and systems may miss what an autonomous AI agent does during a live session. Ultra Security is making that gap the case for its new runtime controls, arguing that agents’ access to tools, services and company data requires oversight while they act—not just checks on network traffic or system settings.
The Birmingham, Michigan, startup emerged from stealth on October 21, announcing a $10 million seed round alongside its agent-security platform. Ultra says its software sits between AI agents and the Model Context Protocol (MCP) servers and connectors they use. The company says each tool call, data read and prompt request passes through the platform, giving organizations a way to observe and control those interactions as they happen.
Policies apply to each tool call
Ultra describes four functions: guardrails, governance, observability and connectivity. Its guardrails apply plain-language policies to tool calls, with controls intended to block access to secrets and credentials, redact personally identifiable information, protect health information, limit abuse and prevent data from one connector being written to another. The company says policies can be set for an organization, a team or an individual device.
Governance controls which agents, connectors and tools a person or team can use. Ultra says it logs and traces agent-to-tool interactions for later review, with anomaly detection that can surface alerts. Its connectivity layer is described as a universal bridge between agents and connectors, designed to work across vendors and models without per-client setup.
The approach puts Ultra in the path of agent activity, rather than relying only on a retrospective record. The announcement presents that as a contrast with web application firewalls and API gateways, which Ultra says were designed for traffic between people and systems, and cloud posture tools that check configurations at a point in time. The release does not include independent performance testing or detail how the product behaves when an agent or connector bypasses the path Ultra monitors.
Local-first deployment and compliance claims
Ultra says the platform runs on a customer’s own machine by default, keeping tool calls, payloads and audit trails local, while offering a remote-hosted option. The company also says it holds SOC 2 Type II, ISO 27001 and ISO 42001 certifications and conforms to AARM Extended. It says it aligns with the NIST AI Risk Management Framework and Agentic Trust Controls, has completed a FedRAMP 20x Class B audit with a third-party assessment organization, and is seeking FedRAMP certification.
Those details come from Ultra’s launch announcement, which does not name customers or publish product pricing. The company says it worked with customers and design partners during a year in stealth, but gives no deployment totals or customer examples in the release.
A $10 million seed round backs the launch
Investors in the seed round include Ludlow Ventures, Hidden Capital, Entrée Capital, Detroit Venture Partners, firedrop and FDVC, alongside individual investors from security and software-as-a-service circles. Ultra was founded in 2025 by Chase Lee, whom the announcement identifies as the former founder and chief executive of Trustpage, acquired by Vanta. The release says access to the platform can be requested through the company’s website.
Ultra’s launch joins a growing set of products aimed at controlling agent activity inside business systems. Its pitch is narrower than a claim to secure every part of an AI deployment: place policies and logging at the agent’s tool interactions, then let organizations decide which actions to allow.