Anthropic details how it measures Claude's wokeness (www.theverge.com)

Anthropic has announced new measures to ensure its AI chatbot, Claude, remains "politically even-handed." This initiative follows President Donald Trump's recent executive order advocating for “unbiased” AI models, prompting AI companies to reassess their algorithms. Anthropic's blog highlights efforts to equip Claude with a system prompt designed to avoid unsolicited political opinions, uphold factual accuracy, and represent multiple perspectives, a move intended to address growing concerns about AI bias. To operationalize this goal, Anthropic employs reinforcement learning techniques that reward the model for producing responses aligned with predefined traits promoting neutrality. Claude has already shown promising results, achieving scores of 95% and 94% in political even-handedness for its versions Sonnet 4.5 and Opus 4.1, respectively—significantly outperforming competitors like Meta's Llama 4 and OpenAI's GPT-5. The significance of this development lies in its potential to shift AI model standards towards inclusivity in political discourse, thereby fostering user autonomy in forming judgments without skewed influences.