In a startling display of AI-driven cybersecurity, three Indian-origin researchers from Hacktron AI successfully breached OpenAI systems using Anthropic’s Claude models. Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini managed to gain access to employee accounts and a private GitHub environment in under 72 hours.
The operation was remarkably efficient, costing less than 3,000 dollars. By leveraging the reasoning capabilities of a rival LLM, the team identified and exploited vulnerabilities that bypassed traditional security layers, highlighting a new era of automated penetration testing.
This incident underscores a growing paradox in the industry: the same tools designed to enhance productivity are now being weaponized to dismantle the security of AI giants. As LLMs become more sophisticated, the battle between AI-powered attacks and defenses will likely define the next frontier of cybersecurity.
Source: Read original article