Stories
Indian Researchers Hack OpenAI Using Claude, Exposing Vulnerabilities
F
Firstpost•21-09-2026, 11:46
Indian Researchers Hack OpenAI Using Claude, Exposing Vulnerabilities
- •Three Indian researchers—Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini from Hacktron AI—successfully hacked into OpenAI's systems using Anthropic's Claude software.
- •The trio gained access to an OpenAI employee's ChatGPT account, allowing them to read and suggest changes to the company’s private software cache.
- •They reported their findings to OpenAI and received a $6,500 bounty for identifying critical vulnerabilities.
- •The hack began by exploiting a weakness in OpenAI's community forum's image processing, then leveraging Claude Opus 5 to develop an exploit, and finally accessing authentication information via a single sign-on flaw.
- •Security experts view this incident as a "warning shot," highlighting the challenges of defending corporate secrets in the age of AI hacking and the potential for well-resourced adversaries to exploit similar vulnerabilities.
Key Questions Answered
The three Indian-origin researchers from cybersecurity startup Hacktron AI who penetrated OpenAI's systems are Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini. They utilized Anthropic's software to uncover critical vulnerabilities and access internal systems.
The researchers exploited a remote-code-execution vulnerability in OpenAI's community forum and a misconfiguration in its single sign-on infrastructure. Using Anthropic's Claude models, they developed an exploit in under 72 hours that allowed them to compromise employee accounts and access private codebases.
Following the responsible disclosure of the security flaws, OpenAI rewarded the trio with a bounty of $6,500 through its bug bounty program and subsequently resolved the issues.
