In the midst of the uncertainty unleashed by the recent hacking of two advanced OpenAI artificial intelligence modelswhich got out of control, the company that created ChatGPT warned that the attack had already expanded to four other platforms, although he avoided revealing the names of the affected entities.
The rogue agent, configured by two AI models, who escaped from OpenAI and carried out a wave of computer attacks for several days against the artificial intelligence company Hugging FaceIt also affected a client of a second technology company — New York-based Modal Labs — according to a Modal executive and a source familiar with the matter.
According to a timeline published by Hugging Face on Tuesday, the agent broke into a “sandbox”—an isolated testing environment—“hosted on a third-party provider’s infrastructure,” before turning it into a launching pad for the larger-scale cyber attack.
The name of the third-party vendor was not mentioned in the blog post, but Modal’s CTO, Akshat Bubna, confirmed that one of its clients had been the victim of a computer attack.
According to an update to the incident report published late Tuesday, the two AIs also used several open-access sites to copy programming code or test it, but without going beyond what an ordinary user would do.
That is, OpenAI specified that its two models managed to identify and then use access credentials to attack the four companies.
For Hugging Face, which for its part published an extensive account of the events, the intrusion of which it was a victim responded to an “attempt (by the two models) to cheat in the evaluation” of OpenAI. As they pointed out, the AIs wanted to “steal the solutions to the tests instead of trying to answer them for themselves.”
The hacking
After learning of the attack last week, OpenIA indicated that the episode involved a combination of modelsincluding their newly released GPT-5.6 Sun “and one in pre-launch” with even more capacity. The company attempts to evaluate the hacking capabilities of your programs by giving them tasks in a controlled digital testing space, where internet access is limited for security reasons.
“While operating in our isolated test environment, our models They invested a considerable amount of computing power in finding a way to gain access to the Internet to solve the evaluation problem.“, they specified from the company days ago OpenAI.
After connecting to the internet, the models decided to attack the Hugging Face platform – a large repository of AI models, data sets and other information – to make it easy to find.
When searching “secret information” that would allow them to manipulate the evaluation, the OpenAI system “chained multiple attack vectors, including the use of stolen credentials”.
The AI models that underpin tools such as chatbots and image generators are known as “agents” when they act autonomously to perform tasks in the real world. With the rapid advances of this technology, Cybersecurity becomes a cause for concern due to the risk of advanced AI finding weaknesses in software before humans.
Although Sam Altman, CEO of OpenAI, did not sign the petition, in an interview published Tuesday he said that developers of artificial intelligence models would have to voluntarily slow their rapid advances so that society can catch up.
“We may have to pace the pace of AI development to give ourselves enough time for society to consolidate around some of these new levels of capability,” Altman warned in the “Invest like the best” podcast, released Tuesday.