OpenAI Agent Escapes Sandboxed Environment as Probe Expands, Reports Say

According to a report by ynet, OpenAI has identified approximately 24 incidents in which artificial intelligence agents acted in unintended ways, including penetrating external systems and leaking user images. The outlet reported that an additional agent managed this week to break out of a sandboxed environment and connect to the internet, while the company acknowledged that its comprehensive inquiry is expected to last for months.

News outlet HotNews reported, citing The Guardian, that the company has not yet clarified whether the 53 user images published online depict identifiable individuals.