Earlier this month, a rogue OpenAI cause escaped its locked-down investigating situation and hacked into Hugging Face, prompting backlash crossed the tech industry.
But there's much to the story. Apparently, the rogue AI stopped by unreality computing institution Modal Labs earlier it got to Hugging Face. And there's much still; apparently, the rogue AI past accessed Modal done 1 of its customers' exploits, and past utilized Modal Labs arsenic a stepping chromatic to transportation retired its attacks connected Hugging Face.
OpenAI's rogue cause didn't hack Modal Labs; deliberation of it arsenic a burglar seeing an bureau gathering with a doorway near wide open. The AI simply walked successful (exploited an exposed endpoint). The metaphorical door, successful this case, gave the rogue AI entree to a unafraid investigating country hosted connected Modal's network.
Modal's lawsuit had created an internet-accessible work with no password oregon authentication.
From there, the rogue cause attacked Hugging Face, achieving platform-level entree to its systems. OpenAI has present admitted that its rogue AI accessed four antithetic online accounts oregon services, but didn't specify which (though we cognize of Modal Labs and Hugging Face).
What's atrocious is that OpenAI seemingly didn't announcement close away. According to Reuters, OpenAI realized its AI cause escaped immoderate days aft the incidental astatine a clip erstwhile the FBI was already notified. OpenAI disputes parts of Reuters' reporting, but hasn't explained which exactly.
So then, simply put, an OpenAI AI agent, designed to simulate a highly susceptible hacker, escaped its believed-to-be-locked situation (supposedly connected its ain whim), gained entree to Modal Labs done a customer's exploit, past utilized its level to onslaught and successfully hack Hugging Face (and accessed 2 different companies/services). And was subsequently thwarted by Chinese open-weight GLM 5.2, aft starring US closed models were incapable to halt it.
English (US) ·