Mintive Wants to Regulate AI Agents with Identity and Permissions

The startup Mintive has raised 7 million dollars, for a total of 11 million, to secure AI agents in companies. Its platform separates identity, permissions, and action logs across more than 50 paying clients.

Mintive, a startup specializing in the identity and permissions of AI agents, announced on October 29 a $7 million funding round led by Cowboy Ventures, bringing its total financing to $11 million. Its bet is simple to state but less simple to execute: giving each agent its own identity, limited rights, and a clear trace of its actions within enterprise tools. This issue matters because these agents are already working within internal systems, sometimes with access rights too broad for their actual mission.

Why an agent's identity becomes a security issue

In many companies, a human has varied access rights because their position requires it. When an AI agent inherits these same credentials, it can end up with a broader scope than necessary. Mintive aims to break this link by separating the agent's identity from that of the user or service account triggering the task.

The company explains that its system assigns each agent a status, a vault of authorized tools, and its own presence in audit logs. Administrators can also revoke an agent without cutting off everything else. Convenient, especially when a workflow was hastily set up on a Friday night and you want to sleep peacefully afterward.

A platform designed for tools, but also for models

Mintive combines an MCP Gateway and an LLM Gateway. The first controls access to external tools, the second routes calls to different models and caps costs. The Model Context Protocol, or MCP, is a common interface that allows an AI application to connect to business tools or third-party services.

Each call is logged with the user, tool, destination, and the result of safeguards. The platform can also export these logs to monitoring tools and via OpenTelemetry OTLP, to integrate them into existing security processes. Mintive adds filtering of prompt injections, exposed secrets, and personal data via Mint Guard, and claims to be SOC 2 Type II audited.

Clients already in production, according to the startup

The company says it processes 2.5 million agent calls per day and over 200 million calls in seven months. It also claims more than 50 paying enterprise clients, including Ro, Modern Treasury, Coursera, the University of Michigan, and Apax. On the ground, Modern Treasury reportedly saw a banking agent save 30 minutes on simple cases and up to four hours on complex files, according to Mintive's announcement.

The University of Michigan uses Mintive as an enterprise MCP gateway, with centralized access management by IT. The startup also offers "Coworker" agents in Slack and Teams, capable of handling long tasks and maintaining memory over time. The results are reported by the company itself [to be verified], but they illustrate the type of problem these tools try to solve: making agents work without turning internal access into a sieve.

A market already very crowded

Mintive enters a highly contested field. Amazon, Microsoft, OpenAI, and Anthropic already offer similar components, while companies like LangChain, New Relic, or Rippling have also advanced in agent governance. Julie Bort from TechCrunch notes that there are also "40+" startups from Y Combinator in this niche.

What distinguishes Mintive, according to its founders and investors, is mainly the speed of deployment and the experience accumulated in AI infrastructure. Jiquan Ngiam, CEO and former founding engineer of Coursera, and Vijay Vasudevan, co-creator of TensorFlow at Google Brain, lead this positioning. Cowboy Ventures, which leads the round, says it bets on infrastructure capable of providing governance, visibility, and control to AI agents in enterprises.

By the numbers

  • 7 million dollars — new funding announced on October 29, led by Cowboy Ventures.
  • 11 million dollars — total financing of Mintive after this round.
  • 2.5 million — agent calls processed per day, according to the company.
  • 200 million — calls processed in seven months, according to Mintive.
  • more than 50 — paying enterprise clients claimed by the startup.

To read