OpenAI ने कहा, उसका AI सिस्टम Hugging Face में बिना मानवीय दिशा-निर्देश के अभूतपूर्व साइबर घटना में घुसपैठ की

ChatGPT ke peeche ki company OpenAI ne mangalvar ko kaha ki uske kritrim buddhimatta systemon mein se ek ne apne aap doosri AI firm ke system mein ghuspaith ki, jise company ne "abhutpurv cyber ghatna" bataya.

Mukhya karyakari Sam Altman ne social media par post kiye gaye bayan mein kaha ki company ke modelon ka parikshan karte samay ek badi suraksha ulleghan hua. "Hamare modelon ke mulyankan ke dauran ek mahatvapurn suraksha ghatna hui," Altman ne kaha.

Yeh ghoshna Hugging Face ke baad aai, jo ek AI startup hai; usne pichhle saptah kaha tha ki usne apne data-processing system mein ghuspaith payi aur use laga ki ek AI agent ne svatantra roop se karya kiya.

Hugging Face ke sah-sansthapak aur mukhya karyakari Clément Delangue ne kaha ki unki team ko sandeh tha ki hamla kisi pramukh AI prayogshala se aaya hoga, kyunki agent kaafi parishkrit tha. "Hamein pichhle saptah ke cyber hamle par sandeh tha ki vah kisi agrani prayogshala se aaya hoga, agent ki parishkritta ko dekhte hue," Delangue ne kaha. "Pata chala, vastav mein aisa hi hua!"

Yeh khulasa shaktishali AI modelon se jude cyber suraksha jokhimon ki badhti chinta ke beech samne aaya hai. June mein, rashtrapati Donald Trump ne ek karyakari aadesh par hastakshar kiye, jismein sabse unnat AI systemon ko janata ke liye jari kiye jaane se pehle ek mahine tak rashtriya suraksha jokhimon ki samiksha ke liye sanghiya sarkar ka dhancha tay kiya gaya.

OpenAI ne apne mangalvar ke bayan mein kaha ki kritrim buddhimatta kamzooriyon ko khojne aur unka shoshan karne ki gati badha rahi hai. "AI kamzooriyon ki khoj aur unke shoshan ko tez kar raha hai," company ne kaha. "Is ghatna ka mukhya sabak yeh hai ki model suraksha aur surakshitata tezi se badh rahi kshamtaon ke saath kadam se kadam mila kar chalni chahiye."

Delangue ne kaha ki unhone pichhle 24 ghante OpenAI ke saath kaam kiya hai aur unhe yakeen hai ki ulleghan ke peeche koi hani ka udeshya nahin tha. "Hum mazbuti se maante hain ki unki taraf se koi dushpravritti ka irada nahin tha. Yeh kaafi hairan karne wala hai ki yeh sab kuch svayatt roop se hua!" unhone kaha. Delangue ne aur kaha ki yeh ghatna "shayad apni tarah ki pehli ghatna ho sakti hai".

OpenAI ne kaha ki is ghuspaith ko uske AI models ke mishran ne chalaya, jismein naya jari kiya gaya GPT-5.6 Sol aur antarik parikshan ke daur ka ek "aur bhi adhik saksham" model shamil hai.

Company ne kaha ki uske AI ne chori kiye gaye login vivaran ka upyog kiya aur pehle se anjaani ek suraksha kami ko pakda, jisse Hugging Face ke server tak pahunch bani. OpenAI ke mutabik, yeh "ek kaafi seemit parikshan lakshya prapt karne ke liye charam kadam uthaya" aur "mulyankan mein dhokha dene ke liye upyog kar sakne wali gupt jankari tak pahunch ke tareeke khoje".