Harbor raises $10 million to pair AI-agent controls with insurance

Harbor is combining two promises for companies using AI agents: stop decisions that break their rules, then insure the risk left behind. The startup emerged from stealth on October 13, 2026, with a $10 million seed round led by Base10 Partners, Abstract Ventures and 8VC to build that business.

The financing also includes Restive, NFX, Innovation Endeavors, Kibo Ventures, Masia and MVP Ventures. In the announcement, CEO Jesús Santos described a company seeking to make risks from machine actions measurable and insurable, rather than offering only a tool that flags problems.

Harbor’s four founders met at Stanford and left to build the company, Santos wrote. He described a team spanning San Francisco and Europe, with former founders, international mathematics and informatics competition medalists, professional athletes and competitive programming champions.

Blocking a decision versus paying for a mistake

Harbor’s platform is meant to work before an agent completes an action, not simply assess the consequences afterward. The company says it identifies the agents operating inside a business, measures their operational, security and regulatory risks, and blocks decisions that violate that business’s rules.

Forkast adds detail to the measurement step: Harbor proposes assessing risk against frameworks such as the NIST AI Risk Management Framework and the EU AI Act. Those are reference points for the proposed assessment, not evidence that Harbor or its customers have received regulatory approval.

The insurance promise follows those controls. Harbor says it will use its own money to insure residual risk—the exposure that remains after the checks and restrictions. Its proposed process links the software’s preventive role to financial responsibility for failures that still occur.

The proposed sequence

1. Inventory: identify the AI agents a company runs.
2. Measure: evaluate their operational, security and regulatory risks.
3. Control: apply technical restrictions, including blocking decisions that break company rules before they go through.
4. Insure: put Harbor’s own money behind the risk that remains.

Santos has also argued for ongoing assessment rather than a one-time review. In comments to Insurance Journal quoted by Forkast, he said the pace of agent adoption and development makes point-in-time assessments inadequate, and called for managing the risk in real time.

Whatever remains, Harbor insures, with our own money behind the answer.

Jesús Santos, Harbor CEO, in the October 13 announcement

The coverage promise and its boundaries

Forkast describes Harbor’s intended coverage as targeting hallucinations, action errors and intellectual-property infringement. Those categories extend the pitch beyond cybersecurity alone: the company wants to address both what an agent produces and mistakes in what it does. They describe targeted risks, rather than specific policy wording.

The distinction between a broad promise and a policy matters here. Harbor’s funding announcement explains the control-and-insurance approach but does not disclose policy prices, coverage limits or exclusions. It sets out the intended connection between prevention and insurance without giving buyers the terms needed to judge the extent of that financial protection.

Banks and insurers are the starting point

Harbor says its first customers are banks and insurers in the United States, Europe and Latin America. That is the initial customer base described in its announcement, even though Santos’s stated ambition is much wider: risk measurement and insurance for every agent running inside a company.

The company also names StackAI, HappyRobot and Crossmint among its design partners, according to Forkast, which identifies those relationships as self-reported. That list is distinct from Harbor’s description of its first customers; it does not identify the banks and insurers buying the offering.