And OpenAI didn't try to silently degrade performance of their top model if its (extremely sensitive) safety sensors went off ...
Anthropic is the silver lining keeping p(doom) below 1.0