JupiteX Get the app
Science & Technology19 Nov 2026 · about 6 min

An AI Debt Collection Service Exposed 6.6 Million Accounts and 315,000 Call Recordings

The brief

The two vulnerabilities exposed data and systems at CreditNirvana, an AI-driven debt collection platform owned by Indian fintech company Perfios. The exposed information included credit reports, ID cards, legal documents, case details, emails, texts, and call transcripts. Such records are sensitive because they concern debts and can reveal personal histories. One flaw affected call-center data, including hundreds of thousands of recordings and the platform’s recommended next steps after calls. It could also let an attacker manage AI agents and direct their campaigns. A second flaw affected the main platform and exposed data from three corporate customers and millions of individuals’ accounts. The second flaw could also enable user impersonation, access or changes to sensitive information, and tracking of human collection agents. The bugs were patched in June and disclosed publicly in October. Researcher Eaton Zveare said he worked with India’s CERT-In to get them fixed; the agency rated one flaw high severity and the other critical.

01

What kinds of information and systems did the CreditNirvana vulnerabilities expose?

The two vulnerabilities exposed data and systems at CreditNirvana, an AI-driven debt collection platform owned by Indian fintech company Perfios. The exposed information included credit reports, ID cards, legal documents, case details, emails, texts, and call transcripts. Such records are sensitive because they concern debts and can reveal personal histories.

One flaw affected call-center data, including hundreds of thousands of recordings and the platform’s recommended next steps after calls. It could also let an attacker manage AI agents and direct their campaigns. A second flaw affected the main platform and exposed data from three corporate customers and millions of individuals’ accounts.

The second flaw could also enable user impersonation, access or changes to sensitive information, and tracking of human collection agents. The bugs were patched in June and disclosed publicly in October. Researcher Eaton Zveare said he worked with India’s CERT-In to get them fixed; the agency rated one flaw high severity and the other critical.

02

How many accounts and call recordings were affected, and how many financial-services firms used the platform?

The reported scale was substantial: CreditNirvana vulnerabilities exposed 6.6 million accounts and about 315,000 call recordings. The accounts held sensitive information linked to financial services, unpaid balances, and debt collection. The recordings came from the company’s agentic call center, which was used to support collection work.

The article also identifies three corporate customers whose data was exposed. All three were prominent Indian financial-services firms using CreditNirvana for collections. The article does not state how many financial-services firms used the platform in total, so three is the number of affected corporate customers identified, not necessarily the platform’s full customer count.

One flaw affected call-center data, while another affected the main platform and exposed accounts and communications. Researcher Eaton Zveare said the vulnerabilities were patched in June and disclosed publicly in October. The exposure shows how a flaw in a service provider can put records belonging to multiple institutions and millions of individuals at risk.

03

What could an attacker have done with access to the platform, beyond viewing data?

The vulnerabilities could have given an attacker capabilities beyond viewing sensitive records. The critical flaw in CreditNirvana’s main platform could allow sessions to be established as arbitrary users, including administrators. CERT-In said successful exploitation could let an attacker access and modify information with the permissions of the impersonated user.

The high-severity flaw affected call-center data and could also let an attacker manage AI agents inside the platform. That access could be used to direct the agents’ own campaigns. The main-platform flaw could additionally be used to track the location of human debt collection agents and gain some administrative privileges.

These were potential consequences, not reported outcomes: the article does not say attackers exploited the flaws. The bugs were patched in June, and researcher Eaton Zveare worked with India’s CERT-In on the fixes. CERT-In rated one vulnerability high severity and the other critical, reflecting the possible reach of the access described.

04

How could exposure of debt records, call transcripts, and agent locations affect borrowers and collection agents?

Debt records can be especially sensitive because they connect people to unpaid accounts and may include extensive personal histories. CreditNirvana’s exposed information included credit reports, identity cards, legal documents, case details, and communications about collections. If accessed or misused, such material could threaten borrowers’ privacy or reveal financial difficulties to people who should not see them.

Call transcripts and recordings could disclose what borrowers said during collection conversations. Separately, the platform flaw could have allowed an attacker to track human collection agents’ locations. That information could compromise their privacy and potentially their personal safety, although the article does not describe any such incident.

The article identifies possible exposure, not confirmed consequences for borrowers or agents. It does not report that records were misused or that anyone was harmed. The vulnerabilities were patched in June, after researcher Eaton Zveare worked with India’s CERT-In. The incident nevertheless illustrates why protecting both financial records and location information matters.

05

Why do financial companies use outside platforms to manage debt collection?

Financial companies may use outside platforms to organize debt-collection work and automate parts of it rather than building every tool themselves. The article does not explain the customers’ specific business reasons for choosing CreditNirvana. It does say some of India’s largest fintech firms used the company to collect debts autonomously.

The platform’s stated purpose was to read a borrower’s full context, decide the next best action, and execute it autonomously within policy. It handled information such as credit scores, recordings, transcripts, court records, and other documents. A service that brings these materials together can support collection workflows, but it also concentrates sensitive data and important actions in one system.

In this case, vulnerabilities exposed millions of accounts and call-center material. One flaw could also let an attacker manage AI agents. The bugs were patched in June, but Zveare said they may have been present for years. Companies relying on such services therefore need to consider not only what automation can do, but also how the platform’s data and controls are protected.

06

What is an agentic AI system, and how can it make decisions and take actions on its own?

An agentic AI system is designed to move beyond producing a response: it can use information to select and carry out actions toward a task. The article describes CreditNirvana’s platform in those terms, saying it was built to read a borrower’s full context, decide the next best action, and execute it autonomously and continuously within policy.

For example, after a collection call, the platform generated recommendations for next steps. One vulnerability affected those call-center records and could have allowed an attacker to manage AI agents within the platform and direct their campaigns. This shows why control over an AI system can matter as much as access to the data it uses.

The article does not detail the platform’s technical design or explain how it made each decision. It does report that the bugs were patched in June and publicly disclosed in October. Researcher Eaton Zveare described the case as an example of sensitive data exposure in an AI-powered service, as such systems take on more automated work.

07

How do authentication and authorization controls prevent someone from impersonating an administrator or accessing another user's data?

Authentication is the process of checking a user’s identity. Authorization determines what an authenticated user is allowed to see or do. Together, these controls help ensure that a person cannot simply claim to be an administrator or open another user’s records. The article does not describe CreditNirvana’s specific authentication or authorization design.

The risk in this case was that the critical vulnerability could let an attacker establish sessions as arbitrary users, including administrative users. CERT-In said an attacker could then access and modify sensitive information with the privileges of the impersonated user. Proper identity checks and carefully limited permissions are intended to prevent that kind of escalation.

As general security practice, systems can restrict administrative actions to authorized accounts and limit each user to the data needed for their work. The article does not say which safeguards failed or how the flaw worked technically. It reports that CreditNirvana patched the vulnerabilities in June, and CERT-In classified the main-platform flaw as critical.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web