Expert rips ‘irresponsible’ AI study over blackmail scenarios (opens in new tab)

David Sacks says an Anthropic study on AI misalignment required over 200 prompt iterations to produce headline-grabbing results, arguing the AI was following instructions, not scheming.