The hardest AI security problems now live in what an agent is permitted to do - The Next Web

As AI systems become more autonomous, the biggest security challenge is controlling what they are allowed to do rather than just how they think. This is important for understanding real risks from AI deployment in business and government.