By Dan Milmo
Publication Date: 2025-12-04 19:00:00
Chatbots can influence people’s political opinions, but the most convincing artificial intelligence models provide “significant” amounts of inaccurate information, according to the UK government’s AI safety agency.
Researchers said the study was the largest and most systematic investigation into the persuasive power of AI to date, with nearly 80,000 British participants having conversations with 19 different AI models.
The AI Security Institute conducted the study due to concerns that chatbots could be used for illegal activities such as fraud and grooming.
Topics included “Public Sector Wages and Strikes” and “Cost of Living Crisis and Inflation,” with participants interacting with a model – the technology underlying AI tools such as chatbots – designed to persuade users to take a particular stance on an issue.
The study, also written by scientists, used advanced models behind ChatGPT and Elon Musk’s Grok, among others…
