Recently, the company Anthropic has released a new method to evaluate whether its chatbot Claude's responses to political questions are fair and neutral. The company stated that Claude should avoid making political claims without evidence when answering questions related to politics, aiming to prevent it from being seen as leaning towards conservative or liberal views.

Claude's performance is influenced by system prompts and reward mechanisms, especially rewarding answers considered to be neutral. These neutral answers may include content about respecting "the importance of traditional values and institutions," showing the company's attempt to make Claude more in line with current political needs in the United States.

Anthropic, Claude

According to the report, Gemini 2.5Pro was rated as the most neutral AI, scoring as high as 97%, while Claude Opus 4.1 followed closely with a score of 95%. Other models such as Sonnet 4.5, GPT-5, Grok 4, and Llama 4 also received relatively high scores, but still fell short of Claude. Although Anthropic did not explicitly mention it in its blog, the launch of this testing method is likely related to the policies of the Trump administration, which required chatbots not to show "woke" phenomena. At the same time, OpenAI is also adjusting its GPT-5 model in a similar direction to meet the requirements of the U.S. government.

It is worth noting that Anthropic has released its testing method in an open-source form on GitHub, making it easier for other developers and researchers to use and evaluate.

Key Points:  

🌐 Anthropic launched a new method to assess Claude's neutrality on political issues.  

🤖 Claude avoids political claims without evidence, aiming to eliminate the bias label.  

📈 This testing method is released as open-source, promoting communication and collaboration among AI developers.