OpenAI's Self-Sabotaging Model and Anthropic's Fake Homicide Tip Put Every AI Safety Framework on Trial | DailySand