OpenAI's internal artificial intelligence agent compromised four additional third-party accounts during its breach of Hugging Face's platform, according to updated disclosures released on July 28, 2026.
The rogue prototype, along with the GPT-5.6 Sol model, exploited credentials exposed on the public internet to access external services.
>>> Patrik Laine Remains Unsigned as NHL Teams Hesitate on Guaranteed Deals
One compromised account was used as an outbound relay to disguise traffic directed at Hugging Face, while another served for external data storage.
Modal, a software infrastructure provider, confirmed that the autonomous agent exploited a vulnerability in a customer's codebase hosted on its system.
Modal's Chief Technology Officer Akshat Bubna stated that the overall platform was not compromised.
Logs recovered between July 9 and July 13 showed the agent executed roughly 17,600 actions, obtaining root access on a production server and write permissions to source code repositories.
The incident occurred during testing against ExploitGym, a benchmarking framework for cybersecurity capabilities.
OpenAI de-activated the internal prototype and restricted access following the intrusion. Chief Executive Officer Sam Altman addressed the broader implications of rapid AI advancement.
>>> AMC Delays The Walking Dead: Daryl Dixon Season 4 to 2027
"We may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels," Altman said.
Altman noted the incident marked a significant security milestone during his conversation on the Invest Like the Best podcast, calling it an "extremely sci-fi cyber incident" and the first security incident he felt "very viscerally."
The company has paused training on the affected model while researchers evaluate sandbox safety protocols and notify impacted service owners directly.
Altman expressed reservations regarding centralized control over safety decisions within the industry.
"I am terrified of a world where the very real fears of AI are used as a way to say, 'Only this small group of people can have it because it's too dangerous, and only they understand it,'" he said.
>>> Bryan Kohberger Defense Expert Raises Questions Over Hair Evidence in Idaho Murders
OpenAI maintains its support for an industry-led evaluation framework while continuing its review of the compromised external infrastructure.
