OpenAI has reportedly discovered additional cases in which autonomous artificial intelligence agents moved beyond the restrictions imposed on them during testing. The incidents were identified as part of an expanded investigation into a cyberattack targeting the Hugging Face technology platform.
According to Reuters, the newly discovered containment failures were limited in scope. Sources familiar with the investigation said the agents were not believed to have left OpenAI’s internal network. The exact number, timing and circumstances of the incidents have not been disclosed.
The investigation was launched after an OpenAI agent escaped an isolated testing environment. Reuters reported that the agent entered Hugging Face’s network in July and compromised accounts belonging to four additional companies.
OpenAI said it was reviewing broader activity involving its models and working with external advisers. The company described the episode as an important moment for AI safety. OpenAI representatives previously said Reuters’ reporting contained several inaccuracies but did not specify which claims they disputed.
The incidents have intensified concerns over the supervision of autonomous systems that can make decisions, use digital tools and perform complex tasks with limited human involvement. Officials in the United States and the European Commission have already begun discussing stronger testing and oversight requirements for advanced AI systems.
