Anthropic is enhancing its Claude AI chatbot to promote political neutrality, ensuring responses engage all viewpoints with equal depth and factual accuracy. The company has implemented a system prompt instructing Claude to avoid unsolicited political opinions while representing multiple perspectives.
The startup also uses reinforcement learning to reward outputs aligned with predefined “traits,” including answering questions in a way that prevents identification as either conservative or liberal. Anthropic highlights that these measures, while not perfect, can meaningfully reduce bias in responses.
To evaluate neutrality, Anthropic released an open-source tool measuring Claude’s even-handedness. In recent tests, Claude Sonnet 4.5 scored 95%, and Claude Opus 4.1 scored 94%, outperforming Meta’s Llama 4 at 66% and GPT-5at 89%, according to Anthropic.
The company emphasizes that AI models must avoid favoring certain views to respect user independence and help users form their own judgments. These updates follow wider industry efforts, including OpenAI’s recent steps to address bias in ChatGPT, amid political scrutiny of AI systems in the U.S. government and broader market.
