AI Misinformation Rate Doubles in 2025, NewsGuard Finds

Generative AI systems are becoming more responsive but also more unreliable, according to a new NewsGuard audit of 10 leading tools. The study found that in August 2025, AI models repeated false claims 35% of the time, nearly double the 18% recorded in August 2024. 

The findings underscore a tradeoff: as chatbots integrated real-time web search, their non-response rates dropped from 31% to zero. Instead of declining to answer, they now attempt to respond to every query. However, this responsiveness has made them far more likely to spread misinformation. 

Key findings include: 

• False claims doubled in one year — from 18% (2024) to 35% (2025). 

• Non-response rates fell to 0%, as chatbots no longer decline to answer. 

• Disinformation sources exploited — malign actors, including Russian operations, are seeding falsehoods online that chatbots surface as credible. 

• Content farms and low-engagement sites are increasingly treated as legitimate sources. 

The shift reflects a core tension in AI development: balancing helpfulness vs. accuracy. By moving away from refusals, systems now scrape a polluted online ecosystem where unreliable data is difficult to filter out. NewsGuard’s report suggests this has inadvertently created a pipeline for disinformation, allowing propaganda to be repeated and amplified by AI tools. 

The audit raises concerns for businesses, governments, and media organizations relying on AI for news insights and real-time analysis. The findings also call into question the industry’s assurances that newer models are safer and more reliable. 

As demand grows for timely AI-powered news summaries, developers may face mounting pressure to strengthen fact-checking protocols, invest in source reliability detection, and introduce transparency measures before misinformation undermines trust in generative AI tools. 

 

Source: 

https://www.newsguardtech.com/ai-monitor/august-2025-ai-false-claim-monitor/