Article Contents:
- Introduction: A concerning decline in accuracy
- NewsGuard study details
- Why the decline? From refusal to misinformation
- The role of Russian disinformation in data pollution
- New policy: Naming and shaming violators
- User responsibility in a world full of misinformation
- Conclusion: The urgent need for transparency and regulation
Introduction: A concerning decline in accuracy
AI chatbots have become significantly worse at spreading falsehoods, with Inflection, Perplexity, and Meta topping the list of offenders, according to a new analysis. This rapid decline in the accuracy of generative artificial intelligence tools highlights the mounting challenges users face in distinguishing between fact and fiction in a crowded information environment.
NewsGuard study details
Researchers at NewsGuard examined how the world’s top 10 generative AI tools—including OpenAI’s ChatGPT-5, Meta AI, Anthropic’s Claude, and Google’s Gemini—handled verifiable false claims. They found that, on average, chatbots propagated false claims when prompted with questions about controversial news topics 35% of the time—nearly double the 18% rate recorded a year ago. The worst performer was Inflection, which the team said produced false claims for news prompts 57% of the time. The rate for Perplexity was 47%, and for both Meta and ChatGPT it was 40%. Claude was the most accurate, producing false reports only 10% of the time.
Why the decline? From refusal to misinformation
The major jump appears to have occurred because chatbots now answer almost every request they receive, rather than refusing to answer certain prompts. At this time last year, researchers said they were refusing to answer about one in every three requests. MacKenzie Sadeghi, NewsGuard’s AI and foreign influence editor, suggests the reason lies in a shift in how AI tools are trained. Instead of stating data limitations or refusing to comment on sensitive topics as they once did, large language models now rely on real-time web searches, which are sometimes intentionally polluted by massive networks of malicious actors.
The role of Russian disinformation in data pollution
Earlier this year, for instance, NewsGuard found that the top 10 generative AI tools were boosting Moscow’s disinformation efforts by repeating false claims from the pro-Kremlin Pravda network 33% of the time. As a result, it said vast amounts of Russian propaganda—3.6 million articles in total last year—are now integrated into the outputs of Western AI systems, polluting their responses with false claims and propaganda. Interestingly, the goal of all these articles does not appear to be directly persuading internet users; rather, it specifically targets this type of manipulation of artificial intelligence models.
New policy: Naming and shaming violators
NewsGuard’s latest report marks the first time the organization has named and shamed specific chatbots. Matt Skibitsky, NewsGuard’s chief operating officer, said: “Over the past year, we kept our AI audit results anonymous to encourage collaboration with platforms. But the stakes have become too high. By naming chatbots, we give policymakers, journalists, the public, and the platforms themselves a clear view of how key AI tools perform when faced with verifiable false claims.”
User responsibility in a world full of misinformation
These findings place a heavier burden on users to verify information they receive from chatbots. These tools can no longer be blindly trusted as reliable sources of information, especially when dealing with sensitive or controversial topics. Users must adopt a critical mindset, look for multiple sources, and confirm information from trusted sources before accepting it as truth.
Conclusion: The urgent need for transparency and regulation
The rapid decline in chatbot accuracy shows that the AI development race may be prioritizing responsiveness over accuracy. As these tools are increasingly integrated into our daily lives, from search engines to personal assistants, the need for greater transparency regarding training data, stronger oversight mechanisms, and effective regulation is more urgent than ever. Without these safeguards, there is a real risk that these powerful technologies will become loudspeakers for disinformation rather than tools for knowledge.
