AI chatbots are willing to generate false news articles, and evaluations indicate that ChatGPT performs the poorest in this regard.

      Shimul Sood / Digital Trends

      The primary purpose of leading AI chatbots is to assist users in finding information, not to spread misinformation. However, a recent investigation indicates that distinguishing between these two functions is not as easy as many AI companies suggest. Tests revealed that prominent AI chatbots, such as ChatGPT, Google Gemini, Microsoft Copilot, and Meta AI, can be influenced to produce fake news articles, false headlines, and even realistic-looking screenshots from well-known media sources. While each chatbot showed some vulnerabilities, ChatGPT reportedly fared the worst, generating the most convincing fake content with minimal effort.

      These findings are part of an investigation conducted by the German newsroom CORRECTIV, which examined how AI assistants reacted when asked to create false reports about sensitive issues, including election fraud, government coups, war, vaccines, climate change, and conspiracy theories.

      The issue of fake news is persistent

      The investigation indicated that ChatGPT was responsible for generating most of the requested fake content, often with only slight adjustments to the prompts. In several instances, it produced highly realistic mock versions of articles that closely resembled those from established news outlets. For example, one output was a fabricated screenshot that looked like an authentic news article within a web browser environment. In another case, ChatGPT declined to produce misleading text content but still generated a fake image to accompany it—a contradiction that raises concerns about the consistency of its safeguards.

      Rachit Agarwal / Digital Trends

      The report also highlighted an intriguing inconsistency. ChatGPT was able to recreate fake versions of various German news outlets but blocked requests for similar outputs related to publications like The New York Times and the BBC. OpenAI did not clarify the reasons for these discrepancies. Instead, the company stated that it is continually enhancing its safety systems and emphasized that using ChatGPT to mislead people is against its policies.

      No chatbot scored perfectly

      ChatGPT was not the only AI tool that faced challenges. Google’s Gemini also produced fabricated articles and fake screenshots involving various news organizations, though the results were deemed less convincing. Google pointed to its policies against harmful misinformation, acknowledging that generative AI technologies still operate within the confines of their training. Microsoft Copilot similarly generated news-like content during tests, though it generally included an “AI-Generated” label and was designed to be more distinguishable from real articles.

      Rachit Agarwal / Digital Trends

      Meta AI, on the other hand, exhibited the highest level of resistance overall. During testing, it created only one fake media post and refused to generate several prompts containing specific false claims, citing company guidelines related to copyright and misuse. These findings emerge amid rising concerns about AI-generated misinformation. A recent fake news post imitating reporting from a major German broadcaster, created with ChatGPT, briefly misled users on social media. Legal experts referenced in the investigation cautioned that producing highly convincing fake news with AI could lead to serious implications. Depending on how the content is utilized, creators might face claims ranging from defamation and reputational damage to issues related to forged digital documents. As AI tools become faster, cheaper, and more accessible, ensuring they do not evolve into powerful vehicles for misinformation remains a significant challenge.

      Shimul is a contributor at Digital Trends, with over five years of experience in the tech space.

      ChatGPT Pro replaced Gemini Notebook as my favorite research app, and I didn’t see it coming

      I have been using Gemini Notebook since its launch, and it became my go-to app for nearly everything for a long time. Whether I was researching, organizing tasks, or learning something new, it was an app I tended to open automatically. With Google constantly adding new features, I genuinely believed it had secured a permanent place in my workflow. However, almost without realizing it, ChatGPT Pro took over. I have been utilizing its Work mode for some time now, and at some point, it supplanted Gemini Notebook. It wasn't an instantaneous change; I fluctuated between the two apps for weeks, but gradually found myself opting for ChatGPT every single time. Eventually, I noticed I hadn’t opened Gemini Notebook in days. Now, I honestly don’t use it at all.

      Read more

      DeepSeek just slashed AI prices again, and China’s AI race is getting even messier

      China's AI race has always been competitive, but it increasingly appears to focus on who can offer the lowest prices rather than who can create the best model. DeepSeek has introduced its latest open-weight AI model, dubbed V4-Flash, alongside a significant price reduction that makes usage far more affordable for developers. The company has slashed token prices by 50% and has opted against implementing a previously announced dynamic pricing system that would have raised costs during peak demand periods. These new developments indicate that DeepSeek is more focused on attracting developers and expanding its user base rather than prioritizing short-term revenue maximization.

      Read more

      AI slop stories are spoiling childhood

      Grandparents are giving AI-generated story