UN-backed scientists called for stronger AI safeguards after agents breached HuggingFace, exposing risks of autonomous systems acting beyond human control.