Arva AI Raises $20M to Sell Compliance Agents to U.S. Banks
Arva AI raised $20 million in Series A funding led by Arthur Ventures, says CEO Rhim Shah, targeting the 40 to 50 largest U.S. banks with agents that automate sanctions and alert reviews.

Why it matters
- Arva AI raised $20 million in Series A funding led by Arthur Ventures, reported by Axios on November 17, 2026
- CEO Rhim Shah said customers include Wise and two large U.S. banks, with targets set at the 40 to 50 largest U.S. banks
- Arva reports more than 1 million reviews processed monthly and 92% alert resolution in a production deployment
- An October 2, 2025 case study showed a 78% cut in alerts reaching human analysts, backlog cleared in 45 days, false-negative rate below 1% over 90 days
- Previous $3 million round announced January 14, 2025, led by Gradient with Y Combinator, Amino Capital and Olive Tree Capital
Arva AI raised $20 million in Series A funding led by Arthur Ventures, Axios reported on November 17, 2026, citing chief executive Rhim Shah. The money funds an expansion of the New York sales team and new AI research, including hiring and model training.
Shah said "customers include Wise and two large U.S. banks," and that Arva "has expanded from serving smaller fintechs to targeting roughly the 40 to 50 largest U.S. banks."
The stakes here are narrower than the category suggests. Banks still send sanctions and suspicious-transaction reviews to human analysts, and the list of buyers for Arva is short. Winning approval at a small set of large institutions comes first. Expanding inside those accounts comes second.
What does Arva's software actually do?
Arva's agents research businesses and individuals, review documents, and investigate alerts tied to sanctions and suspicious transactions. They also recommend improvements to a bank's own review process. Shah said customers increasingly use the agents to automate reviews after testing their performance.
Under the hood, the research architecture builds specialized reasoning layers on top of general-purpose language models, handling tasks such as matching identities, tracing beneficial owners, and processing adverse media.
Arva was founded in 2024 by Shah and chief technology officer Oli Wales. The company operates from London and New York.
How does Arva charge for it?
The company prices on usage. That model creates room to grow inside an existing customer as Arva takes on more categories of work.
- Pricing is based on usage, with revenue rising as Arva handles more work inside each institution.
- Initial deployments are relatively standardized, while adding workflows requires more engineering with the customer.
- Bank vendor approval is difficult, and Shah said that friction can create an opening to sell additional services after deployment.
- The Series A proceeds expand the New York sales team and fund AI research, including hiring and model training.
On the approval process, Shah described bank vendor approval as difficult, but said it can create an opening to sell additional services once a deployment lands.
How much work is Arva actually doing?
On its website, Arva reports more than 1 million reviews processed each month and 92% alert resolution across a production deployment. Those figures are company-published rather than independently audited.
They still tell you the operating scale Arva wants to point at, as it pursues larger bank customers and broader deployments within institutions it already serves.
What is AgentCore, and who keeps control?
AgentCore is Arva's continuous-learning system. It identifies recurring review patterns and proposes evidence-backed recommendations for changing how reviews happen.
The pipeline behind it looks like this:
- Analyst adjustments, escalations and corrections get anonymized into evaluation datasets.
- Agents are graded against expert-labelled datasets and adversarial cases before deployment.
- Subsequent changes are regression-tested against those datasets.
- AgentCore validates proposals through historical replay and tracks performance after deployment.
- Customers decide which recommendations to apply, inside their own sandboxed environments.
That last point matters most. The system proposes; the bank disposes.
Where did Arva come from?
Arva announced a $3 million round on January 14, 2025, led by Gradient, with Y Combinator, Amino Capital, Olive Tree Capital and fintech angels participating.
That earlier money went into automating Know Your Business verification, the compliance process of assessing business customers against risks such as fraud and money laundering. The platform pulled fragmented registry, website, social-media and document data together and turned it into business risk assessments.
Gradient and Y Combinator also participated in the Series A.
What did the October 2025 bank case show?
An October 2, 2025 case study describes an unnamed large U.S. bank. Its rules-based transaction-monitoring system generated thousands of alerts daily, and human reviewers could not keep up.
Arva did not replace that system. It added a decisioning layer between alert generation and human review.
- Alerts were classified for automatic resolution, elevation or escalation.
- Each decision carried an audit explanation.
- Alerts reaching human analysts fell 78%.
- The backlog cleared within 45 days.
- The false-negative rate stayed below 1% during a 90-day quality-assurance period.
- The bank then widened Arva's remit into case-narrative drafting and suspicious activity report enrichment.
The last item is the commercial signal. One workflow won the contract. More workflows followed.
What has to go right next?
The Series A addresses two connected problems: landing approvals at a relatively small set of large banks, then turning each first deployment into additional workflows. Arva's deployment model makes the constraint explicit — broader automation has to stay explainable, validated and under customer control, which is exactly what regulators and bank risk teams tend to demand before they let agents near their alerts.
Source: unite.ai
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
708 articles
Related articles
- Vijil CEO Vin Sharma: AI Agents Need Trust Infrastructure Before Autonomy
- Novo launches AI banking tools with approval-based controls
- Community banks are deliberately slow-walking AI adoption
- Over 40% of Agentic AI Projects Face Cancellation by 2027
- redalpine Closes $250M RAC VIII Fund for European AI-Science Startups