Beltic Raises $7.3M for Know-Your-Agent Rules in AI Commerce

Identity checks have long asked one quiet question: is a person really sitting at the keyboard? A San Francisco startup that just stepped out of stealth says that question no longer covers the way software buys things. Beltic has raised $7.3 million in seed funding, led by Norwest, to build what it calls know-your-agent checks — rules that decide what an AI agent may do on a business's website before the agent can act.

The financing brings Beltic's total raised to roughly $8.8 million since its founding in 2025, according to the company's interview with This Week in Fintech. Alongside Norwest, Restive Ventures, Oxford Seed Fund and Collide Capital participated in the round. A group of angel investors from companies including Coinbase, Legora and Ramp also joined.

Fraud tools were built for humans

Beltic was founded in early 2025 by Isha Bhatnagar and Mike Allan, two founders who had been working on opposite sides of the same problem: letting legitimate transactions through while blocking fraud. Bhatnagar spent nearly six years at Coinbase, where her work included customer onboarding, fraud prevention and compliance products. Allan founded Atar, a banking-as-a-service provider in Brazil that was acquired by insurer Porto Seguro in 2021. The two began discussing how payment and identity systems would hold up once software started acting for customers, and Allan moved from Brazil to San Francisco to build the company.

The pair's central claim is that the checks businesses rely on still assume a human operator. Fraud systems look for device fingerprints, selfies, mouse movements and typing speed — evidence an agent almost certainly cannot provide. A legitimate agent would therefore fail checks designed for people, leaving the business stuck between blocking everything and waving through software it cannot evaluate. In interviews, the founders described that impasse as the problem Beltic exists to solve.

Know your agent, not just your customer

Beltic's product is designed to identify visits from agents and then verify who is behind them. The software checks each proposed action against rules set by the business. A company might let an agent browse flight options for a customer but require another confirmation before the agent can book or pay. When an agent drifts from the user's intent or attempts something risky, Beltic can flag the behavior and require the agent to reconfirm, according to the founders' description of the product.

The founders draw a distinction between their approach and pure agent-identity systems. Establishing who sent an agent is useful, they argue, but the harder questions are what the agent is permitted to do and how to respond when it misbehaves. In a product demonstration, Bhatnagar showed a dashboard that distinguishes agent visits from ordinary traffic, including agents that arrive through a website or through a connection to another piece of software.

The need for action-level control is easy to illustrate. When AI assistants began making restaurant reservations on users' behalf, one prominent investor found his account temporarily banned after his assistant overstepped. Even an authorized agent can misunderstand an instruction or make a mistake that cannot easily be undone, and the founders position their rules engine as the layer between the agent's ambition and the business's checkout button.

Data and distribution come from acquisition

Beltic's ability to identify the people and businesses behind agents grew through an acquisition. Shortly after raising about $1.45 million in pre-seed funding in 2025, the startup acquired Verifiet, a company whose co-founder Farhan Afsahi became Beltic's chief technology officer. The founders say the deal gave them a data platform covering more than 500 million entities across multiple countries, used as the starting point for identifying who is behind an agent before assessing what the agent is trying to do.

The company is currently running pilots with unnamed companies. The team numbers nine people, including six engineers, the two founders and a lead focused on finding customers. Bhatnagar told This Week in Fintech that owning the full stack — from the source data to the verification engine — is what lets the product return sub-second results on agent verification.

Why investors are buying the agentic commerce thesis

Norwest principal Jordan Leites said the firm was drawn to Beltic because the startup sits at the intersection of two areas where it holds strong conviction: the modernization of financial infrastructure and the emergence of agentic commerce. He wrote in an email that verification and authorization infrastructure will have to keep pace as money and commerce move in real time, and that the scale of the opportunity, combined with a team that had lived the problem firsthand, made the investment compelling.

The Beltic raise is one thread in a larger rush of capital into agent infrastructure. Supabase raised funding and acquired Turso to build database infrastructure for agent workloads, and startups like Photon AI raised $4.5 million to put conversational agents on messaging platforms. Each of these bets assumes that agents will soon do a meaningful share of everyday transacting — and that the tools guarding those transactions need to be rebuilt around agents rather than people.

The founders say the goal is not to make businesses afraid of agentic traffic but to let them accept it safely. Companies today either block agents outright, as several major platforms have done, or improvise a way to handle them. Beltic is betting that a know-your-agent layer becomes as standard as know-your-customer checks were for online banking: invisible when things work, decisive when they do not. Reporting for this article drew on the exclusive interview with the founders and investors published by This Week in Fintech.