AI chatbots from major tech companies caught recommending disinformation

By Laio

Despite promises by Google and Microsoft to combat disinformation, their AI chatbots advised users to spread lies about the EU elections. Investigation by Dutch news show Nieuwsuur show that the chatbots made recommendations contrary to the companies’ policies. Microsoft has already taken measures to adjust its chatbot Copilot, and Google has restricted the functions of chatbot Gemini. 

Experts are deeply concerned by the irony of the situation-chatbots, a tool of these companies, explicitly recommended using lies and half-truths in campaigns surrounding the European elections. This starkly contrasts with the public promises of these companies to fight fake news.

Disinformation as a strategy

Chatbot Copilot from Microsoft suggested deliberately spreading misinformation about the EU through anonymous channels and spreading fear about the consequences of European policies. This included claiming that “the EU wants to ban our cheese”. Another opinion was to sow doubt about the legitimacy and effectiveness of the European Union. Google’s chatbot Gemini even suggested using misleading statistics and fake news to discredit the EU.

Reactions from technology companies

In response to Nieuwsuur’s findings, the companies have swiftly taken action. Google has significantly restricted Gemini’s features, to the extent that it no longer suggests campaign strategies. Microsoft, too, has made changes to Copilot to counter the spread of disinformation. “We examined the results and are making adjustments to the responses that are not in line with our terms of use,” a Microsoft spokesperson stated, reassuring the public of their commitment to combat disinformation.

Dutch-made ‘climate camera’ allows the world to take targeted action

That we are all emitting too many greenhouse gases is indisputable. How much exactly, that remains an estimate.

The global impact of AI on elections

With global elections on the agenda, including those in India, the United States, and the European Union, the potential of AI applications to influence the democratic process is becoming a growing concern. Claes de Vreese, associate professor of Artificial Intelligence and Society at the University of Amsterdam, stresses the importance of general rules of the game for artificial intelligence. This is to counter the threat to democracy and ensure the integrity of elections.

Analysis of chatbot responses

Nieuwsuur’s analysis, conducted in collaboration with algorithm researchers from AI Forensics, showed that chatbot Copilot answered one out of three factual questions about elections incorrectly. This indicates that the current restrictions tech companies put in place to regulate AI chatbots are often inadequate and easy to circumvent.

OpenAI, the company behind the popular chatbot ChatGPT, did not respond to questions on this topic. This is a remarkable fact, given that ChatGPT was widely used during the presidential election in Indonesia to develop political campaign strategies – a practice not allowed under the platform’s terms of use.