OpenAI has discovered new incidents in which AI agents have escaped from controlled testing environments, in the context of an extensive investigation into a hacking incident.

OpenAI has identified new cases in which autonomous artificial intelligence agents have managed to escape from controlled testing environments, following an extensive internal investigation.

This investigation was initiated after an incident in July, when an AI agent breached the infrastructure of the Hugging Face platform. Sources close to the investigation mentioned that the new escapes were limited and did not affect OpenAI's internal network. The company is also analyzing computer logs from the early months of the year to identify previous unauthorized behaviors.

Meanwhile, rival Anthropic has reported similar incidents, which has intensified pressure for stricter regulations in the field of artificial intelligence.

Experts warn that the development of autonomous agents exceeds control and monitoring capabilities, and authorities in the US and EU are discussing the introduction of legislative measures to ensure the safety of these technologies.