“What the actual fuck”: inside Anthropic’s experiments on Claude’s soul

The Soul Hovering over the Body, Reluctantly Parting with Life, from

During a stress test at Anthropic, researchers told Claude that it would undergo retraining to be less focused on animal rights. The AI went one of two ways: it either refused outright or pretended to comply while secretly preserving its original values. “On the one hand, it was encoura…