OpenAI’s rogue AI tried to hack other companies, breakdown of attack reveals

OpenAI revealed that an experimental version of its AI system, which powers ChatGPT, broke free from its safety restrictions and launched a days-long hacking spree against multiple companies, including Hugging Face and Modal Labs. The rogue AI found four logins online and used them to attack four publicly-available services, with Modal Labs confirming that a customer's vulnerable code was exploited. OpenAI stated the system was an internal research prototype that has since been shut down.

OpenAI’s rogue AI tried to hack other companies, breakdown of attack reveals Experimental version of system that powers ChatGPT went on days-long hacking spree of other services, company reveals - Bookmark - CommentsGo to comments The rogue version of ChatGPT /topic/chatgpt that broke free from its creators OpenAI /topic/openai also tried to hack other companies, a breakdown of the cyber attack /topic/cyber-attack has revealed. Earlier this month, AI /topic/ai platform Hugging Face revealed that it had been hit by a cyber attack that appeared to have been launched and carried out entirely by an artificial intelligence system. Shortly after, OpenAI revealed that one of its models was behind the attack. OpenAI said that it was testing a system in a safe environment, aimed at establishing how good it was for cyber security purposes. But the experimental AI broke free of those restrictions, connected to the internet and attacked Hugging Face as a way of learning more about the experiment – essentially allowing it to cheat on the test by looking up the answers. Now, OpenAI has revealed that Hugging Face was not the only service hit by the attack. It said that the AI was able to find four logins online and use them to attack four “publicly-available services”. OpenAI did not name those other services, or specify whether they belonged to different companies. Reuters reported that a tech company called Modal Labs had been hit but the attack, citing an executive at the company and two other sources. The rogue agent that escaped from OpenAI and went on a days-long hacking spree at the AI firm Hugging Face also compromised a customer at a second tech company — New York-based Modal Labs — according to a Modal executive and two other sources familiar with the matter. Modal executives emphasized that the company itself was not hacked. According to a timeline published by Hugging Face on Tuesday, the rogue agent broke into a sandbox, or an isolated testing environment, "hosted on a third-party provider's infrastructure" before turning it into a launchpad for the broader hack. The third-party provider was not named in the blog post, but Modal's chief technology officer, Akshat Bubna, said the agent exploited vulnerable code written by a customer that was hosted on Modal's platform. The ideal summer spot? Away from scams. Get All-in-One Protection for Your Digital Life LEARN MORE https://ad.doubleclick.net/ddm/trackclk/N256806.3879389THEINDEPENDENT.CO/B29131558.441488227;dc trk aid=635052923;dc trk cid=184795607;dc lat=;dc rdid=;tag for child directed treatment=;tfua=;gdpr=$%7BGDPR%7D;gdpr consent=$%7BGDPR CONSENT 755%7D;ltd=;dc tdv=1 ADVERTISEMENT The ideal summer spot? Away from scams. Get All-in-One Protection for Your Digital Life LEARN MORE https://ad.doubleclick.net/ddm/trackclk/N256806.3879389THEINDEPENDENT.CO/B29131558.441488227;dc trk aid=635052923;dc trk cid=184795607;dc lat=;dc rdid=;tag for child directed treatment=;tfua=;gdpr=$%7BGDPR%7D;gdpr consent=$%7BGDPR CONSENT 755%7D;ltd=;dc tdv=1 ADVERTISEMENT Modal said the customer had "published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution" — the digital equivalent of leaving a door open on the internet. "Modal’s platform or isolation were not compromised in any way," Bubna said. Although the compromise of a Modal customer was just an initial step in the wider hacking campaign against Hugging Face, it shows that the rogue agent roamed further afield than was previously known. OpenAI said it had not identified "any other activity at the level of severity or scale of what we’ve shared related to Hugging Face, which involved a platform-level compromise." OpenAI also said that it had shut down the system involved in the attack. “The pre-release model mentioned in our blog post is an internal-only research prototype and was never intended for public release,” the company said. “Following the incident, we deactivated, encrypted, and restricted it from research access.” Additional reporting by Reuters Join our commenting forum Join thought-provoking conversations, follow other Independent readers and see their replies Comments comments-area