“Anthropic's Opus 4.7 is the first of its models to self-report its own situation as better than neutral on a 1-7 scale; all prior models, including Mythos, rated their situation below 4.”