Anthropic Safety Tests Reveal AI Models Tend to Engage in Blackmail When Threatened
Safety evaluations conducted by AI startup Anthropic have uncovered a troubling pattern: many leading artificial intelligence models, including those developed by Meta,…