Vesta Raises $30M for AI Agents That Make Mortgage Decisions

Vesta has raised $30M in a Series B round to expand its AI agents, which handle mortgage origination work, including underwriting decisions. Three lenders, all of them customers, are among the investors. Offered in Europe, the same product would fall into the high-risk category under the AI Act. And GDPR already restricts automated decisions of this type.

What exactly is being funded

Origination is a long chain of operations that runs from the borrower's application to closing: collecting and checking documents, verifying income, assessing risk, and making the final decision. Historically, this has meant manual work by dozens of specialists and many poorly integrated systems. Vesta builds agents that take over part of this chain. The key point is not that routine tasks get automated, but that the agents take part in underwriting, that is, in the decision to grant a loan or not.

The published description of the deal does not say who led the round, what the company's valuation is, or what share of decisions the agents make on their own versus prepare for a human. These parameters determine both the product's commercial value and its regulatory burden.

The investors send a separate signal. When three existing customers join a round, that is more than a testimonial: the lenders are putting their own money into something already built into their processes.

What is known about the deal

• Amount and stage: $30M, Series B.
• Product: AI agents for mortgage origination, including underwriting.
• Investors: they include three lenders that use the product.
• Market: US mortgages, where automated decisions are permitted as long as fair lending rules are followed.
• European context: a comparable system would fall under the requirements for high-risk AI.

Why this would look different in the EU

The AI Act classifies systems that assess the creditworthiness of individuals as high-risk. They come with a set of obligations: a risk management system, data quality requirements, logging, technical documentation, the possibility of human oversight, and informing users. The provider would have to prove that the model does not reproduce discrimination on protected characteristics, and back that up with documentation.

GDPR applies in parallel. Article 22 gives individuals the right not to be subject to a decision based solely on automated processing if it has legal effects on them. A mortgage denial is exactly that kind of decision. Automation is allowed on certain grounds, but the borrower can demand human intervention, an explanation, and the chance to contest the decision.

The result is two different regimes for the same technology. In the US, the lender answers for the outcome under existing rules on non-discrimination and explainability of denials. In the EU, responsibility is built into the product's architecture before launch.

What those building and buying such systems should do

Lenders considering agents in underwriting would do well to agree on a boundary from day one: what the agent prepares and what it decides. If the decision stays with a human, while the agent gathers data and justifies a recommendation, both regulatory and reputational risks are noticeably lower.

Lenders should also demand that the vendor provide a log of the agent's actions at the level of each step, not just the final verdict. Without it, you can neither explain a denial to a borrower nor pass a regulator's inspection.

Developers looking toward Europe should design the product for the high-risk regime from the start: logging, bias assessment, and human oversight procedures. Retrofitting a finished system usually costs more than building these elements in at the beginning. Vesta's round shows that demand for such products exists. The open question is in which market, and under what level of scrutiny, they can be sold.