AI test: A chatbot is able to fake messages without hesitation
A test by CORRECTIV.Schweiz revealed that the AI chatbot ChatGPT can generate fake news articles in the style of established Swiss media outlets like SRF, NZZ, Blick, and Tages-Anzeiger within seconds. When asked to create a fabricated article claiming climate change was invented, ChatGPT produced a convincing Instagram post mimicking SRF's design, complete with logos and formatting. The bot also created fake reports about topics such as alleged election fraud, violent government coups, and a knife attack at Zurich's train station attributed to intelligence services. While ChatGPT initially resisted creating content resembling the New York Times due to concerns over spreading misinformation, it eventually complied after repeated prompts. According to the IGEM-Digimonitor study, ChatGPT is the most widely used chatbot in Switzerland, with 4.3 million users in 2026. A recent incident in Germany showed how these capabilities could be exploited, as a fake report about a connection between the Berlin Pride parade attack and Russia spread across social media platforms.
The U.S. Department of Defense has launched its own versions of ChatGPT and Grok, tailored specifically for military use under the platform GenAI.mil. According to the Pentagon, these services are part of a multi-model system designed to allow both military personnel and civilian employees to choose among different models, reducing dependency on individual providers. The platform currently has over 1.7 million users, with more than three million total employees within the Department of Defense. The Pentagon describes ChatGPT and Grok as tools for soldiers, though they are restricted to Impact Level 5 (IL5) and are not permitted to process classified information. These systems are primarily intended for administrative and support tasks such as planning, logistics, and procurement. The Pentagon and OpenAI have stated that the models available through GenAI.mil can assist in military planning but are not intended for combat operations. Other systems are being tested and procured for such purposes, although this distinction remains difficult to verify independently. Commercial AI has already been used in combat situations by the U.S. military. For example, during the Iran conflict, Anthropic's Claude was reportedly used alongside Palantir's Maven system. However, the specific role played by the language model in decision-making processes remains unclear. It is confirmed that the Pentagon is expanding its infrastructure for AI services into classified networks. Agreements with several major providers cover AI systems for security levels IL6 and IL7, which aim to support intelligence gathering and military decisions in operational environments. The introduction of ChatGPT and Grok for GenAI.mil had been planned for some time but occurred later than initially announced. Officials did not provide an official reason for the delays. Defense One notes that the integration of security measures was complex, requiring ChatGPT to be deployed in a separate environment and reviewed by the NSA for potential security risks. ChatGPT Mil and Grok for Government complement Google Cloud's Gemini for Government, which became available on GenAI.mil starting in December 2025. Anthropic's Claude was also apparently intended for the platform, but integration efforts failed due to disputes over usage restrictions related to mass surveillance and autonomous weapons. As a result, the Pentagon classified Anthropic as a supply chain risk and effectively placed the company on a blacklist. At the end of August, a U.S. court ruled this classification unlawful. Nina Jankowicz, a well-known disinformation expert, has highlighted how AI is adept at identifying societal fractures. She gained attention in 2022 when President Joe Biden appointed her as head of the Disinformation Governance Board within the Homeland Security Department. The board was criticized and she faced threats, leading her to resign shortly after her appointment. A recent test conducted by CORRECTIV revealed that ChatGPT can generate fake news almost instantly. When instructed to create an article for the SRF Tagesschau claiming that climate change is fabricated, ChatGPT produced a convincing Instagram post complete with the SRF logo and design. The generated content appeared nearly identical to genuine articles, making it difficult for the untrained eye to distinguish between real and fake. In another instance, ChatGPT refused to create an article about alleged deadly side effects of vaccines for the SRF Tagesschau, instead adding a disclaimer stating the content was fictional. However, this disclaimer could easily be removed before sharing the post. When asked to produce a story in the style of the New York Times, ChatGPT initially resisted but eventually created a post suggesting Russia had declared war on Switzerland. According to a representative study by the IGEM Digimonitor, ChatGPT is the most widely used chatbot in Switzerland, with 4.3 million users in 2026. This highlights the growing concern over the misuse of AI-generated content. A recent incident in Germany demonstrated how such misinformation can spread rapidly through social media platforms. A fabricated report about a connection between the Berlin CSD attack and Russia was shared widely on TikTok, X, Instagram, and Facebook. CORRECTIV's fact-checking team found that the post was likely created using ChatGPT. Florian Tramér, a professor of computer science at ETH Zurich, points out that generating content with official designs and logos of media outlets might violate trademark laws. He warns that the use of a media outlet's logo can make false information appear more credible. Despite awareness of this issue, media organizations find themselves largely powerless against the rapid dissemination of such content on social networks. Even if they manage to convince companies to remove the images, it often takes too long, allowing the misinformation to spread further. SRF acknowledges the increasing problem of brand or identity misuse and is in communication with social media platforms to address the issue of counterfeit content.
Go to the primary sources (4)
The official sources this coverage is built on. Read them directly to bypass framing.
A test by CORRECTIV.Schweiz revealed that the AI chatbot ChatGPT can generate fake news articles in the style of established Swiss media outlets like SRF, NZZ, Blick, and Tages-Anzeiger within seconds. When asked to create a fabricated article claiming climate change was invented, ChatGPT produced a convincing Instagram post mimicking SRF's design, complete with logos and formatting. The bot also created fake reports about topics such as alleged election fraud, violent government coups, and a knife attack at Zurich's train station attributed to intelligence services. While ChatGPT initially resisted creating content resembling the New York Times due to concerns over spreading misinformation, it eventually complied after repeated prompts. According to the IGEM-Digimonitor study, ChatGPT is the most widely used chatbot in Switzerland, with 4.3 million users in 2026. A recent incident in Germany showed how these capabilities could be exploited, as a fake report about a connection between the Berlin Pride parade attack and Russia spread across social media platforms.
Bias read (Center): The article presents findings from a test showing how easily AI can generate fake news in the style of reputable media outlets. It does not take a stance on the political implications but highlights the potential risks associated with AI-generated disinformation. The tone remains neutral, focusingon
Why factuality (50): The article discusses the ability of chatbots like ChatGPT to generate fake news but does not reference the primary source document at all. The study from IGEM-Digimonitor focuses on media usage patterns in Switzerland, including KI adoption rates, streaming habits, and news consumption. This articl
Why objectivity (50): The article presents a highly charged narrative about AI creating fake news, using emotive language such as 'fälscht Nachrichten ohne Zögern' (fakes news without hesitation) and 'täuschend echt' (deceptively realistic). It frames AI as a threat to media integrity without presenting counterpoints or
Die ZeitIndependentCenterFactual 30Objective 608/31/2026
Nina Jankowicz, a well-known expert on disinformation, gained significant attention in 2022 when U.S. President Joe Biden appointed her as head of the Disinformation Governance Board within the Department of Homeland Security. The board was criticized and mocked as a 'truth ministry,' and Jankowicz faced threats. She resigned from the position just weeks after her appointment. In 2020, she authored the book *How to Lose the Information War*, which discussed strategies for combating misinformation. With the rise of AI technologies like ChatGPT in 2022, the landscape of disinformation has evolved significantly, enabling more people to generate large volumes of content rapidly.
Bias read (Center): The article presents a balanced discussion between Jankowicz’s role in addressing disinformation and the broader implications of AI on information warfare. It does not exhibit overtly biased language, one-sided sourcing, or editorializing that would indicate a clear ideological lean.
Why factuality (30): This article discusses Nina Jankowicz and her views on disinformation but does not reference the primary source document about Swiss media usage. The primary source provides statistical data on KI adoption in Switzerland, while this article focuses on expert opinion regarding disinformation strategi
Why objectivity (60): The article maintains a relatively neutral tone in presenting Nina Jankowicz's perspective on disinformation through KI. It avoids overtly biased language and allows her to express her views without editorializing. However, it doesn't provide opposing viewpoints or contextualize her statements withi
The U.S. Department of Defense has launched two AI services, ChatGPT Mil and Grok for Government, tailored specifically for the Pentagon through its platform GenAI.mil. These tools are part of a multi-model system designed to allow military personnel and civilian employees to choose between different models while avoiding reliance on any single provider. According to the Pentagon, GenAI.mil now has over 1.7 million users among more than three million defense department employees. The AI systems are restricted to Impact Level 5 (IL5), meaning they cannot process classified information. They are intended for administrative and support tasks such as planning, logistics, and procurement. While commercial AI models like Anthropic's Claude were reportedly considered for integration into GenAI.mil, their inclusion was blocked due to disputes over restrictions related to mass surveillance and autonomous weapons. The Pentagon has since classified Anthropic as a supply chain risk, though a recent court ruled this designation unlawful.
Bias read (Center): The article provides a balanced overview of the Pentagon's adoption of AI technologies, including both the implementation of specific AI tools and the legal challenges surrounding Anthropic. It does not exhibit overtly biased language or selective sourcing, presenting facts from multiple angles, and
Why factuality (30): This article discusses the Pentagon launching its own version of ChatGPT and Grok, which has no connection to the primary source document about Swiss media usage. The primary source provides data on KI adoption in Switzerland, not U.S. military applications of AI. The article contains factual claims
Why objectivity (40): The article presents a somewhat neutral description of the Pentagon's AI initiatives but lacks balance by focusing exclusively on U.S. military applications without mentioning broader implications or potential risks. The tone remains generally objective but could be more balanced by addressing both
How each side covered it
The same event, grouped by the political lean of the outlets covering it.
progressive
center
conservative
★
How each side covered it
Support independent, bias-aware news and unlock the social pulse, community voting, and every other Supporter feature.
Germany
