Key takeaways
Independent security researcher Eaton Zveare published findings on Nov 18, 2026 about two flaws in CreditNirvana, an India-based debt collection platform that his write-up describes as "a Perfios company."
CERT-In's CVE records describe CVE-2026-110474 (rated 8.7, High) as a "hardcoded static API Key in the client side JavaScript bundle" of the Agentic AI Call Centre application, and CVE-2026-110475 (rated 9.3, Critical) as "improper authentication and authorization controls at the platform login service" of the Collections Platform.
According to Zveare, the flaws exposed about 315,000 call recordings and 6.6 million accounts across three lender clients: Aditya Birla Capital Finance, BharatPe and International Asset Reconstruction Company (IARC).
Zveare's timeline says he reported the flaws through CERT-In on May 30, 2026 and that both were fixed on June 2. AccountsRecovery.net reported that neither his write-up nor WIRED's reporting indicates anyone else accessed the data.
Matt Lippl, Chief Product Officer, PayPath | Reviewed by Matt Lippl
Last reviewed: November 19, 2026
The short version. A security researcher reported two flaws in CreditNirvana, an AI-driven collections platform used by lenders in India. One was an API key left in the code of its AI call center; the other let him sign in as an administrator on the main platform. He says the flaws exposed 6.6 million accounts and about 315,000 call recordings, and that they were fixed on June 2, 2026, the day CERT-In confirmed it would act.
Who this applies to
This post is for lenders, collection agencies, servicers and the compliance, security and vendor management staff who review outside technology providers, including AI agent vendors. It reports what the researcher, CERT-In's CVE records, WIRED and AccountsRecovery.net published. It doesn't assess any company's security or say what any law requires.
What the researcher reported
Eaton Zveare published his write-up, "The AI Repo Man: Exploiting Agentic AI Debt Collection Platforms," on Nov 18, 2026. WIRED covered it the same day. He describes CreditNirvana as "a rapidly growing digital-first debt collection company in India," and WIRED reports the platform is owned by the Indian fintech company Perfios.
The AI call center. Zveare writes that CreditNirvana's Agentic AI Call Center had "a hardcoded API key" and that "The Amazon Cognito token returned from the login was not used at all, which is why the login could simply be skipped." CERT-In's CVE record says an attacker could extract that key from client side JavaScript files and "make arbitrary API calls to retrieve call transcripts and recordings." Zveare lists access to "315k call recordings" for Aditya Birla Capital Finance customers, post-call recommendations, tools to "Build and manage AI-powered campaigns," and an AI agent registry. WIRED reports the flaw "could also have been exploited to allow an attacker to manage AI agents within the platform and direct their own agentic campaigns."
The main platform. On the primary platform, Zveare writes that the Keycloak login token was sent in a request header, but "they did not check it! They chose to rely on the userName in the request body instead." He also reports that an unauthenticated request to a users endpoint returned "a list of more than 10k users." CERT-In's CVE record says exploitation could let an attacker "enumerate platform users, establish sessions as arbitrary users including administrative users, access and modify sensitive information with the privileges of the impersonated user."
What was exposed, per the researcher. Zveare reports 3.4 million accounts in the Aditya Birla Capital Finance instance and "an additional 3.2 million" in the BharatPe and IARC instances, "for a total of 6.6 million." He lists ID cards, credit reports, legal documents and communication activity; more than 161,000 settlements; location tracking of field collection agents; and 433 debt collection agencies working on the platform.
The timeline
From Zveare's write-up:
May 30, 2026: reported through India's Computer Emergency Response Team (CERT-In).
June 2, 2026: "CERT-In confirms they will take action. Later in the day, both vulnerabilities are fixed!"
Oct 26, 2026: CERT-In vulnerability note CIVN-2026-0523 and the two CVEs issued.
Nov 18, 2026: write-up published.
Zveare wrote that "Credit Nirvana / Perfios fixed the vulnerabilities the same day they became aware of them." WIRED reported that CreditNirvana and Perfios "did not return WIRED's requests for comment," and that Zveare "is conscious that the bugs were potentially present in CreditNirvana's systems for years." AccountsRecovery.net reported: "Neither his write-up nor Wired's reporting indicates that anyone else accessed the data."
Zveare told WIRED the episode is "an important example of an AI-powered service having a very sensitive data exposure."
The open question for teams that use AI vendors
The reports describe a vendor platform where one flaw reached recorded collection calls and the controls for AI calling campaigns, and another reached account files for three lenders at once. They don't describe how those lenders reviewed the platform before or after the fixes.
That leaves an open question for lenders and agencies that hand collection work to AI vendors: what does a vendor review actually look at when the vendor's AI agents can place calls, store recordings and recommend next steps? Questions that come up in that discussion include:
Who on the lender's side can see when a vendor's AI campaign is created or changed?
Where are call recordings and transcripts stored, and who can reach them?
How would a lender learn that a researcher had reported a flaw in a vendor's platform?
For background on how U.S. banking regulators describe oversight of outside providers, see our Third-Party Risk Management glossary entry.
Related terms
Third-Party Risk Management ; Agentic AI ; Human in the Loop
Questions, answered
What is CreditNirvana?
Zveare's write-up describes it as "a Perfios company" and "a rapidly growing digital-first debt collection company in India." WIRED reports it is owned by the Indian fintech company Perfios.
What were the two flaws?
Per CERT-In's CVE records, CVE-2026-110474 (8.7, High) was a hardcoded API key in the client side code of the Agentic AI Call Centre application, and CVE-2026-110475 (9.3, Critical) was improper authentication and authorization at the Collections Platform login service.
How many accounts and recordings were exposed?
Zveare reports 6.6 million accounts across three clients and about 315,000 call recordings in the AI call center.
Were the flaws fixed?
Zveare's timeline says both were fixed on June 2, 2026. The CVE records list versions before 2.0.0 of the call center application and before 4.8.2 of the Collections Platform as affected.
Did anyone else access the data?
AccountsRecovery.net reported that neither Zveare's write-up nor WIRED's reporting indicates anyone else accessed the data. CreditNirvana and Perfios did not comment to WIRED.
Sources & further reading
CVE-2026-110474, CVE record (CERT-In, published Oct 26, 2026)
CVE-2026-110475, CVE record (CERT-In, published Oct 26, 2026)
For general information only, not legal advice. Sources are linked above. Teams often review their programs with their own compliance team or counsel.
By Matt Lippl, Chief Product Officer, PayPath | Reviewed by Matt Lippl | Last reviewed November 19, 2026