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

A startup says it has found ways to control AI systems that might act unpredictably or harmfully. If real, this touches on safety concerns that matter to the public, though the details are unclear.