Sumsub Targets AI Agent Risk With 3 Controls For Banks, Insurers And Marketplaces

Sumsub on Monday pushed AI Agents toward enterprise readiness by expanding its Know Your Agent product so businesses can control access, actions and required human intervention.

Key Takeaways

• Sumsub expanded its Know Your Agent product to let businesses control access, actions and required human intervention
• The product is meant to verify agents, set permissions and log actions for audit purposes
• Sumsub built its brand in verification and compliance tools and is applying that expertise to machine actors
• The release highlights controls such as read-only scopes, step-up approvals and audit trails intended to help banks, insurers and marketplaces

Sumsub said the product is meant to verify agents, set permissions and log actions for audit purposes, framing AI agent identity as a trust layer for companies that expect software agents to interact with customer data, internal tools and external services.

AI Agents Identity Is Becoming An Access Problem

That framing matters because the industry has spent most of 2026 talking about what agents can do, not what they should be allowed to touch. An agent can summarize documents, query a database or trigger a payment only if another system grants that access.

AI Agents identity tries to solve that missing piece.

Instead of treating every action as if a human clicked a button, a company can assign an agent its own credentials, scope and activity log. That makes the software easier to monitor and easier to shut down when something goes wrong.

For businesses, the risk is simple.

An employee account usually carries broad permissions built for a person. If an agent borrows that account, it can inherit too much power, turning one bad prompt or one broken connector into a real security incident.

The Industry Has Been Building Toward This

The release lands after months of corporate anxiety around AI agents, especially in sectors with strict audit requirements.

Legal and security teams have pushed back on broad rollouts because many early agent products could take actions without clean identity boundaries.

Also Read: One Task Can Take Hundreds Of AI Calls, So Here’s When An Agent Is Actually Worth It

That concern grew as vendors moved from chat windows to action-taking systems. A chatbot that drafts text is one risk.

An agent that can read contracts, open tickets, copy records or move money is another. Identity controls become more important as the software stops advising and starts acting.

Sumsub built its brand in verification and compliance tools, and the company is stretching that expertise into machine actors.

The bet is enterprises will buy agent controls from vendors already versed in identity, fraud and regulated workflows.

Permissions Are The Real Product

AI Agents identity is not a philosophical idea. It is a permissions system that defines the agent, links it to approved services and sets rules for what data or actions sit inside bounds.

Good systems also create logs that show which agent acted, what tool it called and what result came back.

That record matters when businesses need to answer basic questions after a mistake. Was the action triggered by an employee, an automated rule or an agent?

Which dataset did the agent touch? Who approved the workflow?

Without identity, those answers get muddy fast.

The category is still young, and many products will blur identity with workflow automation. Sumsub’s opening move is useful because it names the problem directly.

Businesses do not just need smarter agents, they need agents with names, limits and traceable actions.

How AI Agents Identity Changes Enterprise Rollouts

AI Agents identity gives a software agent its own verified profile, much like an employee badge in a company directory. Sumsub is trying to make that badge carry permissions, approval rules and logs, rather than letting an agent reuse a worker’s login or an overbroad API key.

Sumsub’s release highlights controls such as read-only scopes, step-up approvals and audit trails intended to help banks, insurers and marketplaces let agents do real work without handing them master keys.

If vendor tools start integrating with common enterprise directories and cloud access systems, agent rollouts could speed up. If they stay siloed, companies may keep pilots small.

Read Next: Agent Memory Overtakes Full Context With 16 Tokens In Real FactKG Test