Ultra Security raises $10m to monitor the work of AI agents

Ultra Security makes safety tools for autonomous AI agents. It came out of stealth with $10m in seed capital. The platform sits between agents and the servers of the Model Context Protocol (MCP), a standard for agents to use tools. Each prompt, data retrieval and tool use goes through it. Staff can see and control agent behavior at the time it happens.

How it works

Ultra says it has four functions:

• Guardrails turn plain language policies into live rules that stop access to credentials.
• Governance controls set which agents and tools each user or team can use.
• Logs record each step between an agent and a tool. Anomaly detection gives alerts.
• One setup serves all agents and connectors, for each vendor or model.

The proof

Ultra gave no customer numbers and no test results. It says it has SOC 2 Type II, ISO 27001 and ISO 42001 certifications. It also says it completed a FedRAMP 20x Class B audit.

What is not known

Ultra says that agent use grows faster than tools that monitor it. It says that older tools cannot see the work of agents in a live session. No external group checked these claims.

Insight: Many firms race to hold the agent checkpoint

Ultra checks the tool use of agents. GoDaddy, Infoblox, Google and Cloudflare showed open identity checks for agents, made with IMDA of Singapore. Forter sells control of agents on merchant sites and says purchases by agents rose four times since August 2026. Each firm guards a different gate. It seems to be a race: the gate that firms use first can set the rules for agents.