Metano raises $12M to secure enterprise AI agents
The new platform uses real-time context graphs to ensure AI agents only perform actions that align with their intended purpose.
The short version
- Metano raised $12 million in seed funding led by Sierra Ventures to build security for enterprise AI agents.
- The platform uses a real-time context graph called Hive to correlate tasks, identities, and data access across entire sessions.
- Security teams can monitor agents without code changes, but full inline enforcement for custom agents requires one line of SDK.
- The tool supports deployment on-premises or within private AWS, Azure, and GCP accounts to keep data under customer control.
Why it matters
As companies deploy autonomous AI agents, traditional security tools struggle to distinguish between authorized actions and malicious intent. Metano provides a way for IT teams to set guardrails that follow the agent's logic rather than just its permissions.
Closing the security gap
Metano emerged from stealth on November 19, 2026, to address a specific vulnerability in enterprise AI. While many organizations are beginning to use AI agents, these systems often possess legitimate credentials that allow them to perform a wide range of actions. The problem arises when an agent begins performing tasks that do not match its original purpose or intended goal.
The company aims to bridge the gap between an agent's permissions and its actual behavior. An action might look legitimate on the surface because the agent has the right to do it, but that does not mean the action is helpful or safe for the business. Metano provides a platform to inventory these agents, their skills, and their potential exposures across various environments.
The startup secured $12 million in seed funding to build this infrastructure. The funding round was led by Sierra Ventures and included participation from Correlation Ventures, High Alpha, and 360 ONE. This capital will support the expansion of a platform designed to manage both homegrown and third-party AI tools.
How Hive tracks intent
At the core of the platform is a real-time context graph called Hive. This system connects various data points, including tasks, identities, tools, and data access, across an entire session. By looking at a sequence of actions rather than single isolated events, the platform can judge the underlying intent of an AI agent.
This approach is necessary because older security systems were not built for autonomous systems. CEO Neelima Rustagi noted that traditional tools were not designed for agents that can make decisions and change behavior during operation. Metano's system identifies suspicious activity by checking if a chain of events aligns with what the agent was originally built to do.
The platform also includes a research group called MARS Labs. This group investigates malicious agent skills, vulnerabilities in Model Context Protocol (MCP) connections, and compromised software dependencies. This research helps the platform detect threats not just in live actions, but in the very components used to build the agents.
Deployment and enforcement
Metano offers different levels of integration depending on how much control a company needs. For discovery, detection, and monitoring, the platform requires no code changes across supported surfaces. This allows security teams to gain visibility into their AI environment quickly without disrupting existing workflows.
For organizations that need full inline enforcement on their own custom agents, the process is more involved. Developers must add one line of Metano's SDK at the time of agent initialization. This creates a clear distinction between simply watching what an agent does and actively stopping it from performing unauthorized actions.
The platform is designed to be flexible regarding where data lives. Customers can deploy Metano on-premises or within their own AWS, Azure, or GCP accounts to ensure data remains under their control. It also integrates with standard security tools like Splunk using common schemas like OCSF.
Adoption and evaluation
The platform is built to be accessible for security teams who need to move quickly. Metano claims that security teams can begin using the platform in under 30 minutes. To help with this, they offer a free Eval Edition that provides 30 days of real-time scanning for risky configurations and agent activity.
The detections provided by the platform map to established security frameworks, specifically MITRE ATLAS and the OWASP Top 10 for Agentic Applications. These are standard ways for security professionals to categorize and prioritize the risks they face in their specific industries.
Early adopters are already using the tool to manage risk. Srivatsa Ramarao, the CISO at Sigmoid, stated that the platform helped his company adopt AI while minimizing shadow risks. He noted that it prevents sensitive data leakage and strengthens the protection of enterprise information.
The path forward
Metano's rollout highlights a practical trade-off for modern IT departments. Organizations can start with a low-friction audit mode to see what their agents are doing. From there, they can make a deliberate decision to move into enforcement mode, which requires more active participation from the developers who build the agents.
The company's focus on session-level intent marks a shift in how AI security is handled. Instead of just checking if an agent is allowed to access a folder, the system checks if the agent's current path of actions makes sense for its job description. This level of granularity is becoming essential as agents become more autonomous.
As more companies move toward using AI to handle complex workflows, the ability to verify the 'why' behind an action becomes as important as the 'what.' Metano aims to be the standard for ensuring these autonomous systems remain within the boundaries of corporate policy.
What to watch
- How many enterprises will choose to stay in 'audit mode' versus moving to full SDK enforcement?
- How will the Model Context Protocol (MCP) impact the frequency of new vulnerabilities Metano has to track?
Try it
Check out the OWASP Top 10 for Agentic Applications to see the common security risks associated with AI agents.
Sources and links
- Officialfounders’ biographies metano.ai
- Officialproduct documentation metano.ai
- ReportingUnite.AI: the article we sourced
First reported by Unite.AI. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.