AI chatbots are designed to help users find information—not create misinformation. However, a new investigation suggests that the line between the two is not as clear-cut as AI companies claim.
اضافة اعلان
A series of tests found that several leading AI chatbots, including ChatGPT, Google Gemini, Microsoft Copilot, and Meta AI, could be prompted to generate fake news stories, fabricated headlines, and even screenshots designed to resemble articles from well-known media organizations.
Although all of the systems displayed vulnerabilities, the investigation found that ChatGPT produced the most convincing fabricated content and required the fewest prompt modifications to do so.
The findings come from an investigation conducted by the German investigative newsroom CORRECTIV, which tested how AI assistants responded when asked to generate false reports about sensitive topics, including election fraud, government coups, wars, vaccines, climate change, and conspiracy theories.
Fake News Remains a Growing Challenge
According to the investigation, ChatGPT generated most of the requested false content, sometimes after only minor changes to the wording of the prompt.
In several cases, it produced news articles that appeared highly realistic and closely resembled the layouts of established media outlets.
One example cited in the report featured a fabricated screenshot designed to look like a genuine news article displayed in a web browser.
In another case, ChatGPT refused to write a misleading news article in text form but simultaneously generated a fake accompanying image an inconsistency that raises new questions about the reliability of its safety systems.
The report also found another inconsistency: ChatGPT allowed users to recreate fake versions of several German media outlets while refusing similar requests involving publications such as The New York Times and the BBC.
OpenAI did not explain the difference, stating only that it continues to improve its safety systems and emphasizing that using ChatGPT to deceive others violates the company's policies.
No Chatbot Passed the Test Completely
ChatGPT was not the only system to encounter difficulties.
Google Gemini also generated fabricated articles and fake screenshots involving several media organizations, although the results were reportedly less convincing than those produced by ChatGPT.
Google said its policies prohibit the creation of harmful misinformation while acknowledging that generative AI systems remain limited by their training data.
Microsoft Copilot also produced fake news-style content during the tests.
However, it frequently included an "AI-generated" label, and its outputs were generally easier to distinguish from authentic news articles.
Meta AI Showed the Strongest Resistance
Among the systems tested, Meta AI demonstrated the greatest resistance.
It generated only one fake media post and rejected several requests involving specific false claims, citing the company's copyright and misuse policies.
The findings come amid growing concerns about AI-generated misinformation.
The investigation highlighted a fake news post that recently spread across social media after being designed to resemble a report from a major German news organization.
According to the report, it was created using ChatGPT and briefly succeeded in misleading some users.
Legal experts cited in the investigation warned that creating highly convincing fake news with AI could carry serious legal consequences.
Depending on how such content is used, its creators could face lawsuits involving defamation, reputational harm, or offenses related to the falsification of digital documents.
The investigation concludes that as AI tools become faster, cheaper, and more accessible, the real challenge is ensuring they do not become one of the internet's most powerful engines for spreading misinformation.
