Harbor Raises $10M Seed for AI Agent Risk and Insurance

Harbor's AI-agent software has an audience outside the engineering department: the carrier considering an insurance policy. The San Francisco startup is building a business around making agent exposure measurable, connecting enterprise risk software with insurance for work delegated to agents. Harbor announced a $10M seed round on October 13, 2026, led by Base10, Abstract and 8VC.

CEO Jesus Santos and co-founders Daniel Elechiguerra, Cristian Bartolome and Luis Botin met at Stanford before building Harbor. Their proposed commercial relationship reaches across the engineering team operating an agent, the organization authorizing its actions and the carrier considering the financial exposure. That makes the evidence produced by the software part of a larger purchasing conversation.

Harbor's approach matters because an agent's access can change faster than the institution's understanding of what that access allows. A new tool or wider permission can alter the consequences of the same customer request. Harbor wants risk assessment and controls to follow those changes, with insurance addressing covered losses under an actual policy.

What Harbor's $10M Seed Round Supports

The official funding announcement identifies Base10, Abstract and 8VC as the leads. Restive, NFX, Innovation Endeavors, Kibo Ventures, Masia and MVP Ventures are also named as collaborators. The company has not disclosed a valuation, individual investor allocations or a detailed budget for deploying the $10M.

Harbor's founders bring different experience to the problem. The company's biographies describe Santos's previous work at Morgan Stanley, McKinsey and Crossmint, while Elechiguerra previously worked at McKinsey and Citadel. Bartolome has a Stanford robotics master's degree and prior founder experience; Botin's background includes Stanford mathematics, computer science and AI safety research. These are company-reported biographies, with Santos's CEO role identified in the signed announcement.

The combination is commercially relevant without needing a heroic origin story. Risk software has to inspect technical behavior, while insurance distribution requires a separate understanding of exposure, underwriting and policy boundaries. Harbor is bringing those conversations into one business at a point when agents are being designed to take actions on company systems.

The company says its first customers are banks and insurers in the US, Europe and Latin America. It has not identified those institutions or published revenue, paid-customer counts or independently verified production results. That disclosure establishes the intended buyer and geographic focus while leaving the scale of the business open.

How Harbor Connects Agent Evidence With Risk

Harbor describes an inventory of enterprise agents followed by assessment of their operational, security and regulatory risks. Its product page presents monitoring, static analysis, adversarial testing and runtime guardrails as parts of that approach. Those are company-described capabilities, rather than an independent assessment of their effectiveness in a customer's production environment.

The components examine different parts of the same operating record. Static analysis considers code paths, tools and permissions; monitoring follows trace history and live activity. Adversarial testing probes unsafe behavior and misuse, while guardrails are designed to evaluate an intended action before execution. Together, these records could help an organization understand both the permissions it has granted and the failures those permissions might allow.

Harbor's illustrative payment example shows why the moment before execution is valuable. An agent may have a tool capable of moving a large amount of money even when a customer's operating policy authorizes only smaller payments. The existence of that tool and the permission to use it are facts a risk team needs to examine. Interface examples on the website demonstrate the concept; their session totals and timestamps should not be mistaken for customer performance statistics.

The company's framework also distinguishes technical safeguards from governance responsibilities. A record of what happened has more value when the organization knows who owns the agent, which policy applied and how the decision can be reviewed. Harbor's investment thesis depends on that evidence becoming useful outside the team that generated it.

The Insurance Relationship Behind the Platform

Harbor's legal disclosure identifies Miralta Q, Inc., doing business as Harbor, as an insurance agent and surplus-lines broker. The disclosure says insurance described on the website is issued by surplus-lines carriers. This makes carrier participation and written policy terms part of the product's commercial reality, alongside the software that measures risk.

Coverage depends on the actual contract, exclusions, limits and jurisdiction. Harbor's website describes protection for several kinds of agent-related failure, but the presence of a risk category on a website does not establish that a particular organization's loss is covered. The company also states that coverage is not bound through ordinary website communication without written carrier confirmation. Public carrier identities, policy pricing and a record of issued policies were not established by the sources reviewed.

For a buyer, this introduces a practical sequence of questions. The engineering team needs to understand what Harbor can observe and stop; the risk team needs to understand the resulting exposure. The insurance conversation then has to establish what a carrier will accept, which events qualify and which obligations remain with the customer. Evidence can help those people work together, but it cannot substitute for the agreement between them.

Harbor is consequently proposing more than a new location for an agent dashboard. Its risk measurements would have a commercial audience with different incentives from the product team. An insurer considering potential losses needs evidence that remains interpretable when an agent changes, and a buyer needs a usable explanation of the protection being offered.

What Changes When Agent Permissions Keep Expanding

Harbor's risk framework groups operational, security and regulatory concerns, then scopes assessment around agent architecture, industry and jurisdiction. Mapping a framework to regulations or standards is the company's description of its method; it is not evidence of certification or a guarantee of compliance. The useful idea is that the assessment should reflect the agent and the environment in which it is allowed to act.

That distinction becomes sharper when a company updates a model, adds a tool or expands access to data. A review of the previous configuration may explain yesterday's behavior while leaving the new permission unresolved. Harbor describes reassessment as part of its approach, which could give technical changes a clearer route into the risk and insurance discussion.

The competitive question extends across agent monitoring, safety testing, enterprise governance and insurance distribution. Harbor is attempting to connect their outputs so a company can examine an action and discuss its financial consequences through a shared record. The sources do not establish a market-share position, superior loss performance or a completed new insurance category, and no such result is implied by the financing.

The $10M gives Harbor financing for that ambition without supplying the missing operating history. Its ongoing relationship with a bank or insurer will develop as the customer decides which tasks an agent may take on and which permissions it is prepared to extend. Each extension gives the engineers another configuration to examine and the insurance conversation another exposure to explain.