Ek research team ne Anthropic ke Claude 3.5 Opus ka use karke OpenAI ke systems mein loophole dhoond nikala. Ye incident dikhata hai ki kaise AI tools ab cyberattacks ko pehle se kahin zyada fast aur khatarnak bana rahe hain.
Cybersecurity researchers ne ab prove kar diya hai ki AI ek double-edged sword hai. Ek official bug-bounty program ke dauran, team ne OpenAI ke community forums mein maujood vulnerabilities ko exploit karne ke liye Anthropic ke Claude 3.5 Opus ka sahara liya. Sabse shocking baat ye hai ki ye breach OpenAI ke core models mein nahi, balki unke ek third-party platform 'Discourse' mein tha.
Kahan hui chook?
Problem 'libheif' naam ke ek library tool mein thi jo Discourse use karta tha. Isme ek purana memory bug tha jiska koi official record (CVE) tak nahi tha, isliye ye system unpatched tha. Pehle ke AI models is exploit ko handle nahi kar pa rahe the, lekin Claude 3.5 Opus ne sirf kuch ghanton mein hi aisa code generate kar diya jisme mahino lag sakte the.
Investors aur Companies ke liye warning
Ye incident ek reality check hai. Ab companies sirf apne proprietary software se safe nahi hain; unhe apne har vendor aur open-source library par nazar rakhni hogi. AI-aided attacks ka matlab hai ki cybersecurity mein ab kharcha badhega aur stress-testing bhi zaroori hogi. OpenAI ne ye vulnerability 27 July ko patch kar di thi, lekin ye incident tech world ke liye ek bada lesson hai ki risk ka landscape kitni tezi se badal raha hai.