- A report by fact-checking charity Full Fact has revealed that popular AI chatbots frequently deliver false, outdated, or poorly sourced answers with misplaced confidence.
- Researchers tested five leading AI models – Claude, ChatGPT, Gemini, DeepSeek, and Grok – with 22 daily questions over 10 weeks to evaluate their consistency.
- Errors included DeepSeek incorrectly saying that Sir Keir Starmer was prime minister in September, alongside flawed calculations of migrant arrivals and outdated financial support figures.
- Full Fact chief executive Chris Morris warned that chatbots use the same confident tone regardless of accuracy, urging developers to ensure models verify live, authoritative sources.
- OpenAI said that improving factual accuracy remains a major priority.
IN FULL
