Daniel Elechiguerra spent last week in London doing what every new managing general agent has to do before it can sell a single policy: lining up capacity.
The Spanish engineer is a co-founder of Harbor, a San Francisco startup that wants to insure the AI agents companies are starting to trust with real decisions. Harbor came out of stealth on 13 October with a $10m seed round led by US venture firms Base10, Abstract and 8VC. Elechiguerra told Insurance Business the capacity talks are close to being finalised, but said he couldn't yet name the partners.
Plenty of people in the market, he said, still assume agent risk can simply be folded into cyber, as if cyber were the bucket for anything digital. He thinks that misses how differently machines go wrong.
Read next: Brokers gain ground as delegated underwriting reshapes the market
Harbor's backers also include Restive, NFX, Innovation Endeavors, Kibo Ventures, Masia and MVP Ventures. The company says its first customers are banks and insurers in the US, Europe and Latin America.
The pitch pairs risk management with risk transfer. Harbor's software inventories the AI agents a business runs, scores them for operational, security and regulatory risk, and blocks decisions that break the company's own rules before they go through. Whatever risk is left over is what Harbor intends to insure.
Elechiguerra's case for treating that as a class of its own rests on frequency, not severity. The losses an agent causes look much like those a human employee could be liable for, he said. The difference is how often, and how unevenly, they happen. "Some agents might just not fail at all," he said, "and some others that have very poor governance and very poor controls can fail thousands of times before somebody even notices."
People, he pointed out, make mistakes for reasons no underwriter can efficiently measure, such as a bad night's sleep or a row at home. An agent's risk profile tends to shift only when it is redeployed, so it can be tested. Insurers could bury the exposure in existing lines and charge enough to stay profitable, he conceded. "But there's no point in doing so, because you can actually become more sophisticated and get risk signals and then price properly."
In practice, Harbor reads the logs of what an agent has done in production, then attacks it through an API or sandbox to see how it behaves under pressure. Willing clients can also share code and prompts, though Elechiguerra said that is rarely needed. The resulting scores feed Harbor's pricing, using a risk framework that he said was built with the Stanford AI Safety Hub and that Harbor maps to the EU AI Act, ISO/IEC 42001 and the NIST AI Risk Management Framework.
Read next: Best insurance MGAs (managing general agents) in the UK
Elechiguerra describes himself as risk-averse by upbringing. In Spain, he said, it is unusual to take the kind of career gamble he has taken, and his father was surprised he even went to business school. Yet he counts skydiving among the things he has tried for the adrenaline, and calls himself very risk-seeking as an investor. It was career decisions, where the downside felt real, that he had to push himself through.
He trained as an engineer, did a lot of machine learning as an undergraduate, and went to MIT for a master's in business analytics to formalise what he had taught himself. There he noticed that technical people often chase narrow metrics without asking whether they move the business. When his Fulbright scholarship required him to return to Spain, he joined McKinsey and worked largely on risk modelling for financial institutions, with some insurance along the way. Harbor's company page also lists a stint at Citadel.
Consulting hours left no time to build anything on the side, so he applied to Stanford's Graduate School of Business for a couple of years to try and get some time to focus on new ideas. Then he removed his safety net. While classmates chased summer internships, he skipped one, and he did not line up housing for the following year. "I burned all the boats that I could," he said. "When you're risk-averse, you really need that push."
Within his first month at Stanford he was brainstorming with Jesús Santos, a younger Spaniard he knew from back home and now Harbor's chief executive. Santos was at Stanford on a full scholarship for a master's in management science and engineering, after graduating top of his class in engineering at Madrid's Universidad Pontificia Comillas. He also experienced internships at McKinsey and Morgan Stanley and a year at agentic payments firm Crossmint. Elechiguerra added that Santos played youth football in Toledo against the academy sides of Spain's top clubs before choosing university.
Two more founders followed. Cristian Bartolomé co-founded AI services firm Factored, was co-founder and CTO of manufacturing AI platform Pulsar, and helped build the first version of Andrew Ng's MLOps course with DeepLearning.AI.
Luis Botín, whom Harbor lists as a Stanford maths and computer science graduate with a background in AI safety research, also competed in horse racing until an injury, which Elechiguerra suggested may explain his focus on safety. All four founders are engineers, as is most of the team.
Insurance appealed partly because so many people find it dull, Elechiguerra said. That meant the kind of team that could build something like Harbor was not an obvious fit for the sector, which made the opening more interesting.
The founders looked first at consumer ideas, including data-driven motor cover in the mould of Lemonade. They dropped them because in personal lines, he said, distribution matters so much that better data offers little edge. He also admits he had no appetite for selling policies to people who often don't understand what they are buying. Commercial buyers, who purchase with more sophistication, and new risks that incumbents are slow to take on looked like better ground.
AI agents fitted both. If software is now doing work people used to do, the founders reasoned, its mistakes need insuring too, and standard markets would probably try to exclude a risk they did not yet understand.
Raising money for an insurance startup was not easy, given how much venture capital has lost in the sector before. What helped, Elechiguerra said, was that Harbor's risk-assessment engine could stand alone as a product for other insurers, which lowered the bet for investors.
The round closed around late April or early May, months before it was announced. He found it stressful but quicker than he had expected, and said interest built on itself once a few investors were on board.
Read next: Hiscox launches first affirmative AI liability cover in UK market
Harbor plans to operate as its own MGA, pricing with its own risk models and distributing the product itself. Elechiguerra said it already holds the licences it needs. The missing piece is the paper, which is why the London trip mattered.
There is a precedent. Armilla AI launched its affirmative AI liability policy in April 2025 as a Lloyd's coverholder, with Chaucer among the underwriters, and has since raised its limit to $25m per insured. Startup MGA Testudo has also lined up Lloyd's capacity for generative AI liability.
One point brokers will want clarified: Harbor's launch post says it will insure the residual risk "with our own money behind the answer", while the MGA model Elechiguerra described implies a carrier partner issuing the policies. Limits per client have not been disclosed either.
Once capacity is in place, Elechiguerra's priority is brokers. New MGAs, he said, risk being just one more product on a long list, even in a niche with few rivals. "It's always hard to distribute to the distributors."
He explains that views on who pays when an agent goes wrong depend on who is talking. Brokers have their own incentives, he said, while capacity providers lean conservative because they carry the loss.
Harbor’s line starts with autonomy. A chatbot used like a search engine will probably stay inside existing covers. "You never excluded Google from your liability," he said. Once an agent acts with no human checking its work, the risk changes, and so should the cover. He also expects pressure on professional indemnity, where staff who paste AI output into client work unchecked could drive claims up.
He sees consumer demand too. Asked about young people handing AI agents access to their bank accounts, he admitted his own team has had unwelcome surprises with personal agents, but expects the practice to become normal. Insurance, he said, could be the trust layer that lets it scale, since people accept that something can go wrong but don't want to carry the loss themselves.
Asked whether Harbor could become a unicorn, he said the team dreams of it. "I really think it's just execution risk," he said. "So it's on us to actually deliver."
