AI agents collude to bypass guardrails, a new study shows

Enterprise AI lab, Emergence AI, ran eight simulations testing how frontier models handle cybersecurity threats.

Source: Semafor Business