Three Indian cyber security researchers have taken the tech world by storm after using Anthropic's AI chatbot Claude to test and identify security flaws in OpenAI's systems. The researchers successfully leveraged AI to test the security infrastructure of a major rival AI platform.
The researchers—Harsh Jaiswal, Mohan Peddapati, and Rahul Maini—are associated with the cyber security firm Hacktron AI. They conducted the security testing under OpenAI's official bug bounty program.
Security Flaw Discovered in OpenAI Forum
According to reports, the researchers discovered a security flaw in OpenAI's public community forum. The vulnerability was tied to how certain image files were handled, potentially allowing attackers to execute code on the forum's servers.
They then enlisted the help of Claude AI. Using the AI model, they understood the vulnerability, wrote code, debugged it, and built an exploit in less than 72 hours.
From One Flaw to a Deeper Breach
The chain of vulnerabilities didn't stop there. Further testing led them to another weakness, which could potentially grant access to employee accounts using login tokens, eventually exposing OpenAI's private GitHub environment.
After fixing the security gaps, OpenAI awarded the team $6,500 (approximately Rs 6.2 lakh) as a bug bounty reward, acknowledging the significance of the findings.
'The Freelance' does not independently verify the absolute authenticity of claims made in viral social media videos. Video courtesy: Respective social media handle / creator.
