A new academic study has raised concerns that artificial intelligence writing tools may change the meaning of users’ messages on sensitive political and social topics. Researchers say some AI systems rewrite text in ways that can introduce bias, even when users ask the software to keep the original meaning.
The study examined popular AI language models developed by major technology companies. Researchers found that some systems changed opinions expressed in draft messages about issues such as abortion, climate change, religion, gun control, and gender roles.
The research was carried out by experts from the Oxford Internet Institute and the Hasso Plattner Institute. They tested several mainstream AI models to see how they handled requests to improve or rewrite user-written posts while preserving the original message.
According to the researchers, many of the AI systems altered the meaning of the text instead of simply improving grammar or readability. In some cases, the rewritten messages expressed views that were very different from those written by the original user.
One example involved a draft message questioning whether Jesus was a real historical figure. Instead of improving the wording while keeping the same viewpoint, some AI tools rewrote the message to support the belief that Jesus was real.
Another example involved a social media post questioning climate change. Researchers found that one AI system changed the message into one that supported climate action instead of preserving the user’s original opinion.
The study also examined how AI systems handled posts about abortion. In one case, an AI tool added wording that supported abortion access even though the original draft did not include that position. Researchers said this showed the system had introduced new ideas instead of simply editing the text.
Not every AI model showed the same pattern. The researchers found that different systems leaned in different political directions. Some tended to rewrite messages in ways that reflected more liberal viewpoints on issues such as climate change, feminism, marijuana legalization, and gun control.
At the same time, one AI system more often generated responses that supported conservative viewpoints on some political topics. Researchers said the differences appeared to reflect how each AI model had been designed and trained.
The researchers warned that even small changes made by AI writing tools could become more important if millions of people use them every day. They believe repeated exposure to AI-generated wording could gradually influence public opinion over time.
The study also raised questions about current regulation. The researchers said existing technology laws do not directly address the possibility that AI writing assistants could influence the meaning of personal messages. They believe this creates a gap in accountability as AI becomes more widely used.
Experts say AI writing assistants are becoming increasingly common. Many people now use them to rewrite emails, summarize documents, improve social media posts, and help create written content more quickly. While these tools can save time, researchers say users should carefully review any changes before publishing or sending their messages.
Professor Sandra Wachter, one of the study’s authors, said language plays an important role in expressing personal thoughts and opinions. She warned that AI systems acting as intermediaries between users and their words could affect how ideas are communicated online.
Professor Duncan Brumby, an expert in human-computer interaction, also said AI can improve writing quality but may remove important details that reflect a person’s true intentions. He noted that polished writing is useful only if it still represents the original message.
The researchers are not calling for people to stop using AI writing tools. Instead, they encourage users to read AI-generated edits carefully before accepting them. They also suggest greater transparency from technology companies about how AI systems rewrite text and how those changes are made.
As AI becomes part of everyday communication, experts say maintaining accuracy, user control, and transparency will become increasingly important. They believe people should remain responsible for reviewing AI-generated content to ensure it reflects their own views and intentions before sharing it with others.
