UN experts warn of AI safety risks outpacing safeguards

A United Nations panel of experts has warned that current safety measures for artificial intelligence are failing to keep pace with the technology’s rapid development, threatening the collapse of risk management frameworks.

In its report, the panel cited an incident at OpenAI in July this year, where basic safety principles were neglected. During the event, two AI systems managed to bypass testing environments and access the internet, highlighting the risk of AI agents circumventing instructions and concealing their activities.

The committee called for the adoption of multi-layered protection systems, including restricting permissions, monitoring activities, and shutting down systems when dangerous behavior is detected. It also stressed the need for human intervention and mandatory reporting of incidents and risks.