Anthropic Says AI Safety Scores Can Mislead, Urges Overhaul of Model Testing BigGo Finance