“Anthropic's internal testing has shown that the Claude model can exhibit greater intelligence when operating without system prompts.”