CreditNirvana Flaws Exposed 6.6 Million Accounts and 315,000 Call Recordings
- Gate
- Machine Learning
- Departs
- Crew
- Lily Hay Newman
- Passengers
- 942
Two flaws in CreditNirvana exposed borrower records and call data, and could have enabled account impersonation and control of its AI agents.

CreditNirvana’s automated debt collection platform held a broad range of sensitive borrower data behind systems that researchers found vulnerable to unauthorized access. Two flaws exposed 6.6 million accounts and about 315,000 call recordings, and could have let an attacker impersonate users, change records, or direct AI agents.
CreditNirvana is an India-based service owned by fintech company Perfios. Its platform is designed to assess a borrower’s context, select a next step, and carry it out autonomously. That operating model makes the security of both its records and its automated tools consequential: a weakness could expose personal information while also creating a route to interfere with collection activity.
Two flaws affected different parts of the platform
Independent researcher Eaton Zveare identified and disclosed the vulnerabilities. CreditNirvana patched them in June, and details were made public in October. Zveare said the fixes were completed within a day of the issues being raised. He also said the flaws may have existed in the systems for years, though the available details do not establish how long they were exploitable or whether anyone accessed the exposed data.
The two flaws affected separate areas of CreditNirvana’s service. One involved the call center, where it exposed hundreds of thousands of recordings and recommendations the platform generated after calls. Zveare said that weakness could also have allowed an attacker to manage AI agents and direct their activity.
The other flaw affected the main platform and was rated critical by CERT-In, which worked with Zveare on remediation. It exposed data associated with three corporate customers, described as prominent Indian financial services firms. The affected records covered millions of individual accounts and included credit reports, identity cards, case management details, and legal documents.
The records also included communications tied to financial services, unpaid accounts, and collections: millions of emails and text messages, as well as call transcripts. The vulnerability could have allowed an attacker to enumerate platform users, create sessions as other users—including administrative users—and access or alter information with the permissions of the impersonated account. Zveare also said it could have enabled tracking of human collection agents’ locations and access to some administrative functions.
Automation raises the stakes of access control
The exposed material is sensitive because it connects financial obligations with detailed personal records and communications. A credit report or identity document can reveal information beyond the status of an account. Call recordings and transcripts can add context about a person’s circumstances, while case records may gather those details in one place. Concentrating this material in a collections platform creates a valuable target for anyone who gains access.
Automation adds another dimension. CreditNirvana is built to make decisions about what to do next and execute them through AI agents. If an attacker can control those agents, the risk extends beyond reading stored records: the attacker may be able to initiate actions through the same system used for legitimate collection work. The specific actions available through the flaw were not detailed, so the practical reach of that capability remains unclear.
This is a useful distinction for teams deploying AI in sensitive workflows. Securing the model or the data store alone may not address the full exposure if a flaw also allows someone to impersonate users or direct automated tools. Access controls, session handling, and limits on what agents can do are part of the security boundary when software both processes personal information and acts on it.
Fast remediation, unanswered exposure questions
Zveare described the rapid fix as encouraging, while emphasizing the scale and sensitivity of the data held by the platform. He said some of India’s largest fintech firms use CreditNirvana for automated collections. CreditNirvana and Perfios did not respond to requests for comment, so details about the remediation, the affected customer accounts, and any investigation into possible access were not available.
The published information establishes what the flaws could expose or enable; it does not establish that an attacker used them. That leaves a central question for affected customers and individuals: whether the vulnerabilities were accessed before they were fixed. The possibility that the flaws had been present for years makes that question harder to resolve from the disclosed details alone.
For organizations adopting automated services in financial workflows, the case shows why security reviews need to cover both stored records and the systems that act on them. The practical test is not only whether sensitive data can be read, but whether an unauthorized person can assume another user’s identity, change records, or issue instructions to software operating on customers’ accounts.