Beltic Raises $7.3M to Verify AI Agent Transactions

Beltic has raised a $7.3M Seed led by Norwest to build identity, authorization, and risk infrastructure for AI agents. Restive Ventures, Oxford Seed Fund, and Collide Capital participated in the financing announced on September 30, 2026.

The San Francisco company is emerging from stealth with $8.8M in total disclosed funding, including a prior $1.5M Pre-Seed led by Collide Capital. Beltic is betting that agentic commerce will need more than payment rails and capable models. Businesses must be able to establish who sent an agent, what that agent may do, and whether a particular action fits the principal's authority and the receiving company's policy.

That problem matters because agent transactions compress time. One human with bad credentials may create one suspicious request. An automated system can repeat the same mistake or abuse pattern across thousands of actions before a reviewer understands what happened. Beltic wants to make the chain of identity, delegation, intent, and accountability visible while the transaction is still being decided.

What Happened

Beltic's $7.3M Seed announcement names Norwest as lead investor, with Restive Ventures, Oxford Seed Fund, and Collide Capital participating. Collide Capital also led Beltic's earlier $1.5M Pre-Seed, joined by Latitud Ventures, Positive Ventures, and unnamed angels. The company did not disclose its valuation or the financing terms.

The company says the capital will support its mission to let businesses verify and authorize agents as they begin acting across commerce and payments. Beltic did not provide a detailed hiring plan, geographic expansion target, or budget by product line, so the use of funds should be understood as a broad product and commercialization mandate rather than a disclosed operating plan.

Beltic was founded in 2025 by Isha Bhatnagar, its CEO, and Mike Allan, its COO. Bhatnagar spent six years at Coinbase working on global regulatory experience, identity, onboarding, and early institutional customers. Allan previously founded Atar, a banking-as-a-service company that operated in Brazil and was acquired by Porto Seguro in 2021.

From Credentials to Authority

Most software security begins by asking whether a credential is valid. Agentic systems create a second problem: a valid credential may still leave the receiving business uncertain about the software actor using it, the person or company behind that actor, and the exact action that was delegated. The risk grows when one agent calls another and the chain continues across systems that do not share the same rules.

Beltic's public product presents three layers of control. Agent visibility identifies the agent, operator, and origin behind a request. A policy engine lets a business define which accounts, actions, and resources carry risk. Intent intelligence evaluates what the agent is trying to do, such as searching, ordering, or transferring funds, rather than treating every request as an isolated technical event.

The company calls the broader model Know Your Agent, or KYA. The analogy to KYC is useful but incomplete. A customer is usually verified as a person or business at defined moments. An agent may need identity, principal authorization, capability limits, action context, and an auditable chain of delegation to be evaluated together and in real time.

Why the Founding Team Matters

Beltic's financing is also a bet on people who have worked inside the systems the company is trying to connect. Bhatnagar's Coinbase experience covers regulated onboarding and identity infrastructure. Allan's Atar background covers banking and payment operations in a complex market. CTO Farhan Afsahi co-founded Verifiet, the data-infrastructure company Beltic acquired in 2025.

According to Beltic and Norwest, the Verifiet acquisition gave Beltic direct data connections across global jurisdictions and coverage of more than 500M entities. That scale is company and investor reported, not an independently audited operating metric. It still explains the strategic logic: an agent credential is only as reliable as the verification performed on the person or business whose authority it claims to carry.

Beltic says it owns the data, context, and verification engine underneath its agent layer. That approach is designed to avoid stitching together a series of external checks every time an agent arrives. The commercial question is whether the integrated stack can stay current across jurisdictions and remain fast enough for automated transactions without turning every interaction into a compliance delay.

The Market Is Building the Rails

The surrounding market is already formalizing how agents reach merchants and complete transactions. Visa's Trusted Agent Protocol describes how merchants can recognize approved agents, verify signed messages, and apply controls when an agent arrives with commerce intent. The Agentic Commerce Protocol defines a programmatic checkout interface between buyers, agents, and sellers.

Those standards help requests and credentials move. Beltic is focused on the decision surrounding the request: who is behind the agent, which authority traveled with it, what the agent intends to do, and whether the receiving business should permit the action. The category will likely include payment networks, identity providers, cybersecurity vendors, compliance platforms, and protocol developers because no single layer owns the entire trust problem.

That makes interoperability as important as verification. Businesses will resist an agent identity system that works only inside one model, payment rail, or platform. Beltic says its infrastructure is designed to work across protocols and payment rails, but broad production interoperability remains a claim to prove through deployments.

What the Seed Must Prove

Beltic says it is working with a select group of design partners in agentic commerce and payments. The company does not name those partners, and its public case studies describe customer types rather than identifiable enterprises. The evidence supports early product work and a credible founding thesis, but it does not yet establish repeatable commercial adoption.

The Seed gives Beltic time to turn its verification stack into infrastructure that developers can integrate and risk teams can defend. The product will need to recognize agents without blocking legitimate automation, preserve policy across chains of delegation, and produce records that satisfy both security teams and regulators. Each requirement becomes harder when transaction volume rises and the agent ecosystem remains fragmented.

Norwest is funding Beltic at the moment when agentic commerce is moving from demonstrations toward real payment and workflow infrastructure. The next stage will be measured less by how many agents the system can identify than by whether businesses trust Beltic's decisions enough to let those agents act. Every permitted transaction becomes a test of the identity, authority, and accountability Beltic says it can keep connected.