Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents - TechCrunch

Researchers have developed a method to control AI agents that might act unpredictably or against intended goals. This addresses a real concern about keeping powerful AI systems safe.