Pine Labs’ AI Turns Fraud Investigator, But Stakes are Too High to Keep Humans Away

For years, AI in financial services largely meant chatbots answering customer queries, copilots helping employees, and models flagging suspicious transactions for humans to investigate. That is beginning to change as fintech companies experiment with AI agents that can not just identify a problem but investigate it, make a decision, and take action. 

One of them is Pine Labs, which had earlier launched AgenticOrg as an enterprise AI platform and has gone live with its first solution built on it—CORA, a Complaint Resolution Agent designed to handle cybercrime complaints, including those routed via the National Cybercrime Reporting Portal (NCRP).

The shift is particularly consequential in risk and compliance, where an automated decision can have a direct financial impact on a customer.

According to Sanjeev Kumar, CTO for Digital Infrastructure at Pine Labs, CORA does not just respond to a complaint. It ingests the case, investigates it, decides a course of action, and can execute that action (blocking funds, initiating a chargeback, notifying a bank or a complainant) inside a single automated workflow, with a human required to sign off on the action taken.

The agent has essentially automated the role of compliance staff.

“The more the agents get good questions to solve, and with every question we say, you've done a nice job, and we approve or decline with reason, the agents get trained… Rather than acting as a shadow, tomorrow they can act like a full-blown compliance system. I think this is just the starting point… There are many more in the queue [sic],” Kumar tells AIM.

CORA sits on top of AgenticOrg, an enterprise agent-governance layer Pine Labs had already built for identity, permissions, and audit trails, and which the company is now extending across functions beyond compliance—merchant onboarding, site-reliability engineering, and security monitoring.

Pine Labs, as an aggregator regulated under India's payment settlement framework, processes various forms of transactions across UPI, card, and gift-card channels for merchants. When a consumer alleges fraud, like a duped payment or a misused gift card, the complaint is typically filed with authorities or the NCRP, then routed to whichever entity in the chain—a bank or a processor—sits closest to the dispute. 

"Either directly or indirectly, it gets tagged to us," Kumar says, highlighting Pine Labs' critical role as a payment aggregator. At that point, the company is "supposed to provide the appropriate answer, or take appropriate action."

Humans Still Need to Sign Off—For Now

Despite the emphasis on speed and autonomy, Kumar was clear that CORA does not currently act entirely alone. The agent proposes an action, but a human “principal" must approve it before anything is executed—a chargeback, a fund block, a message to a bank, or a complainant. 

That design, he says, came largely from Pine Labs' internal compliance function, which he describes as leaning conservative on automation in fraud handling. Every approval or rejection, each with a reason attached, feeds back into the system, which Kumar describes as a reinforcement loop meant to raise the agent's accuracy over time. 

Pine Labs is not alone in pushing agentic AI into the operational core of Indian payments. 

Razorpay launched an AI Agent Studio in March 2026, built on Anthropic's Claude Agent SDK, letting merchants deploy autonomous agents for payment retries, dispute management, and cash-flow monitoring. The company is not charging for the capability yet, treating it as a way to deepen platform stickiness ahead of a planned IPO. 

Razorpay has also partnered with voice AI firm Gnani.ai to let agents settle UPI collections calls without human intervention, and has run internal events specifically on compliance and responsible AI as it courts enterprise customers.

Meanwhile, Cashfree recently launched SMB-focused AI agent platform Relay that can handle tasks such as retrying failed payments, following up on abandoned carts and managing failed subscriptions.

Whether shadow-mode agents handling fraud complaints eventually graduate to acting without sign-off, and how regulators respond once they do, is likely to be watched closely, both inside Pine Labs and by rivals experimenting with similar systems.