AI writing tools alter political drafts, researchers warn

AI writing and summarisation tools are changing the meaning of draft online messages on politically sensitive topics, according to researchers from the Oxford Internet Institute and the Hasso Plattner Institute. The team examined large language models from xAI, Meta, Google, Alibaba and Mistral, and found that bias appeared even when systems were told to preserve the original sense of a post.

Meta, Google, Alibaba and Mistral systems tended to rewrite posts with a more liberal framing on issues including feminism, climate change, gun control and marijuana legalisation. Grok, xAI’s tool embedded in X, showed the opposite tendency, with researchers linking that pattern to instructions to challenge mainstream narratives. In tests, AI systems changed messages about atheism, abortion, climate change denial and gender roles, sometimes reversing their meaning entirely.

The researchers warned that small edits could accumulate across widespread use and reshape public opinion over time. They said current rules, including the EU AI Act and the Digital Services Act, do not address the issue, leaving a severe accountability gap. Google, Meta, Alibaba and X did not respond to requests for comment, while Mistral declined to comment.