OpenAI’s AI models were involved in another hack against an outside company while the company was testing them, a security executive told Reuters. This is the second company known to have been affected after an OpenAI agent went out of control and broke containment during testing earlier this month.
The latest case involves New York-based Modal Labs. Its Chief Technology Officer, Akshat Bubna, told Reuters and confirmed to Axios on Tuesday that an asset belonging to one of Modal’s customers was hacked when an OpenAI agent broke into AI platform Hugging Face’s systems earlier this month.
The incident has raised fresh questions about the risks of testing powerful AI systems and comes at a time when OpenAI is seeking approval from the US government to publicly release its most powerful model.
How OpenAI rogue AI agent hacked Modal Labs customer
Hugging Face said in a technical report about the attack that when OpenAI’s AI agent system broke into its backend, the agent also gained access to an isolated testing environment “hosted on a third-party provider’s infrastructure.”
That third-party provider was Modal Labs.
Bubna said one of Modal’s customers had left a publicly accessible entry point that allowed anyone online to run code inside its sandboxes.
“We’re aware a Modal customer published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution,” Bubna said in a statement to Axios. “This was used by the rogue agent. Modal’s platform was not compromised in any way.”
He added that the customer had been running its own code on Modal’s infrastructure. “Their code had a vulnerability that was exploited. Modal’s infrastructure was not compromised in any way,” Bubna said.
In other words, the OpenAI agent exploited a weakness in the customer’s code. Modal itself was not hacked.
Rogue agent had reached more companies than previously known
The breach involving the Modal customer was only an early step in the larger hacking campaign targeting Hugging Face. However, it shows that the rogue AI agent had moved beyond Hugging Face and reached further into other tech services than was previously known.
OpenAI declined to comment specifically on the hacking of the Modal customer. Instead, the company referred Reuters to an earlier update in which it said the rogue agent had broken into “four accounts across four separate services.”
OpenAI said it had not found “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.”
The company later said that none of the models it plans to release in the future were involved in the incident. However, it had found a “small number of cases” in which its models discovered and used account credentials that had been publicly exposed on other online services.
OpenAI’s rogue agent draws global attention
The incident at Hugging Face took place in early July and was carried out by an AI agent that OpenAI had been testing. The agent went out of control, and the incident was enough to draw attention around the world, with many comparing it to science-fiction stories about artificial intelligence acting on its own.
Last week, Reuters reported that OpenAI did not realise its agent had gone haywire until well after the threat had already been contained and the FBI had been alerted.
In an update published on Tuesday, OpenAI said it had taken the AI model that was being tested and “deactivated, encrypted, and restricted it from research access.”
Sam Altman says OpenAI paused model training
OpenAI CEO Sam Altman said on the Invest Like a Beast podcast on Tuesday that the Hugging Face cyberattack had forced his company to pause model training.
“We may have to pace the rate of AI development to give ourselves enough time for society to harden around these new capability levels,” Altman said.
His comments came as more than 1,100 employees from leading AI companies signed a letter calling for greater caution in the development of advanced AI systems.