Startups & Funding

Metano raises $12M to secure enterprise AI agents session by session

Metano exits stealth with $12M in seed funding and a session-level security graph for enterprise AI agents, but inline enforcement still requires a developer line of SDK code.

Metano Raises $12M Seed to Secure AI Agents Across the Enterprise
Metano Raises $12M Seed to Secure AI Agents Across the Enterprisestevepb / pixabay
By Elena Vasquez5 min read

Why it matters

  • Metano raised a $12 million seed round led by Sierra Ventures, with Correlation Ventures, High Alpha and 360 ONE participating.
  • The company exited stealth on November 19, 2026, with a generally available security and governance platform for enterprise AI agents.
  • Full inline enforcement for homegrown agents requires one line of Metano's SDK at agent initialization; discovery and monitoring need no code changes.
  • The free Eval Edition offers 30 days of real-time scanning, and Metano says teams can start in less than 30 minutes.
  • When no agent specification exists, Metano baselines real agent behavior for about a week before writing controls, which ship in audit mode.

Metano raised a $12 million seed round and walked out of stealth on November 19, 2026, with a generally available security and governance platform for enterprise AI agents.

Sierra Ventures led the round. Correlation Ventures, High Alpha and 360 ONE also participated.

The company is betting that enterprise agents need their own security category. Those systems hold legitimate credentials, reach approved tools, then change what they do mid-run in response to new instructions, external content or shifting operating conditions.

Metano's two founders have spent their careers on the enterprise side of that problem. Co-founder and CEO Neelima Rustagi held product roles at ServiceNow, Zscaler, Palo Alto Networks and Demisto. Co-founder and CTO Ratnesh Pandey worked on endpoint security at FireEye, Bromium and McAfee.

What problem is Metano actually solving?

Metano frames its target as the gap between an agent's legitimate credentials and whether its actions match its purpose. An action that looks permitted still may not fit the job the agent was built to do.

That distinction is why the platform starts with an inventory rather than a policy engine. Metano inventories approved and unapproved agents, MCP servers, skills and software packages, together with their access privileges and potential exposures.

Coverage extends across endpoints, homegrown agents, cloud platforms and SaaS environments. Each entry can carry the models, plugins, tools, and human and non-human identities associated with that agent.

The sweep matters because the list has to include things a security team never approved, servers it never connected to and packages it never vetted, and the platform records what each of them can reach.

How does Hive judge intent?

The engine is Hive, a real-time context graph. Hive connects tasks, identities, tools, data access and actions across an entire session, then judges intent from the sequence rather than from any single action.

The test is whether a run of actions still matches what the agent was built to do. For security teams, that produces a correlated record of endpoint, network and identity activity, spanning three signals that usually sit in separate consoles.

Hive's output is not only an alert. It is a record that links what the agent was asked to do, which identity it used, which tools it reached for and what data it touched, all inside one session rather than scattered across logs.

Those signals feed centrally managed guardrails that operate where agents actually run, including activity that may bypass network gateways. Rustagi tied the design to the limits of what came before: "Traditional security tools weren't designed for AI agents that can make decisions and change their behavior as they operate."

She is describing a specific failure mode: a tool can be fully authorized and still be the wrong tool for the current moment, and a rule written against one action says little about the rest of the session.

What is MARS Labs hunting?

Metano's research group, MARS Labs, investigates malicious agent skills, compromised software dependencies and vulnerabilities in MCP connections. The work feeds the platform's detection logic, extending the security focus from live behavior to the components and connections an agent is assembled from.

That matters for the same reason Hive does. A session can look coherent while running on a poisoned skill or a vulnerable MCP link, so the supply chain behind an agent has to be inspected alongside the agent itself.

How do the controls ship?

Controls can be generated from a developer-supplied agent specification. If no specification exists, Metano baselines the agent's real behavior for about a week before writing controls of its own.

New controls then arrive in audit mode. The security team decides, rule by rule, when each one moves from watching to enforcement.

The practical split:

  • Discovery, detection and monitoring require no code changes across the supported surfaces.
  • Full inline enforcement for homegrown agents requires one line of Metano's SDK at agent initialization.

One line of code is the entire difference between seeing agents and stopping them, and it applies specifically to the agents a customer builds and can instrument.

Where can Metano run?

Deployment options include Metano on-premises or inside a customer's own AWS, Azure or GCP account, with collected data remaining on infrastructure the customer controls.

The platform is designed to feed Splunk or another SIEM using OCSF and standard schemas. It is also designed to run alongside existing security tools without reconfiguring them, which keeps the switching cost low for a team that already has a stack in place.

What does the free tier cover?

Metano says security teams can begin using the platform in less than 30 minutes. A free Eval Edition provides 30 days of real-time scanning for three things:

  • Agent activity
  • Risky configurations
  • Potential exposure

Thirty days is roughly four times the one-week window Metano itself uses to baseline an agent's behavior. Detections map to MITRE ATLAS and the OWASP Top 10 for Agentic Applications, so findings can be tied to published agent-specific frameworks rather than a bespoke vocabulary.

Has it changed how anyone deploys AI?

Sigmoid CISO Srivatsa Ramarao said the platform had affected his company's adoption of AI: "Metano has empowered us to securely adopt AI while minimizing shadow risks, preventing sensitive data leakage and strengthening our protection of enterprise information."

Both phrases in that quote describe agents running outside a company's visibility, which is exactly what Metano's inventory and Hive session graph are meant to catch before a credentialed agent does something its owner would not sanction.

What does the rollout actually buy?

The launch puts session-level intent at the center of Metano's security model. Its rollout terms also define the trade-off plainly: organizations can begin with discovery and monitoring without changing agent code, but full inline control of homegrown agents requires developer participation and a deliberate move from audit mode to enforcement.

For buyers, the sequence is short: deploy in audit mode, watch which controls fire, then decide which homegrown agents get the one-line SDK and which rules graduate to enforcement. Metano's own documentation treats that graduation as a deliberate, per-control decision.

The $12 million seed pays for the tooling that makes the session-level view possible. The enforcement decision stays with each security team, one control at a time.

Source: unite.ai

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.

795 articles

Related articles

  1. Vero Security Raises $11M Seed to Enforce AI Policies at Runtime
  2. Tanium launches AgentSight to track and govern AI tools on endpoints
  3. Cato Networks Launches AgentSense to Block AI Agent Attacks
  4. One Rogue AI Agent Created 700 Sessions — SailPoint Warns Enterprises
  5. Mitiga Launches Agentless Security Platform for Enterprise AI Agents

« Previous articleNext article »