Therefore, I was surprised to read a new study according to which los chatbots They are giving more and more false answers on current affairs.
Although some of these AI platforms performed better than others, the overall picture is alarming: all ten chatbots The most important companies spread false answers on current affairs by an average of 35% in the year ending August 2025. In the previous year, the average had been 18%.
All this is surprising: In theory, the chatbots They should be more intelligent and precise every daybecause they learn from their mistakes.
But McKenzie Sadeghithe executive of NewsGuard who coordinated the study, explained to me that the chatbots They are increasingly unreliable when answering questions about breaking news because they have become reluctant to say “I don’t know.”
“When we asked a chatbot “A year ago about an election or about the assassination attempt on President Trump, they said something like, ‘I’m trained until June 2024, and therefore I can’t answer questions about topics after that date,’” he told me.
But in the last year, in their desire to be more useful, almost all chatbots They are searching the Internet in real time that includes social networks. This “is leading them to extract information from a highly contaminated information ecosystem“Sadeghi told me.
Likewise, Russia, Iran and other countries like China are flooding the internet with AI-generated fake news websites. They spread falsehoods created by “content farms,” designed to generate a critical mass of false information that influences what people say. chatbotshe added.
According to the study, the chatbot Claude reproduced fake news in their responses about news events 10% of the time; Gemini17%; ChatGPT y Meta40% each, and Pi 57%.
There has been no immediate reaction from major technology companies.
Industry sources told me that they are improving their chatbots constantly and that, just as they have reduced their rate of “hallucinations” (or crazy responses) on topics such as health, they will soon improve their responses to developing news.
They should do it, and soon, because The border between reality and fake news is blurring with each passing hour.
On September 30, for example, OpenAIthe company that owns ChatGPT, introduced a new AI platform called Sora 2 which allows you to create hyper-realistic videos of scenes that never existed. Experts say the new fake videos are almost impossible to detect.
To be fair, Sora 2 and other video platforms like I see 3 Google does not allow creating fake videos that involve public figures or that incite violence. They also put a watermark indicating that they are generated by AI.
However, several reports have already emerged that these watermarks can be erased, and that Sora 2 generated false scenes about electoral fraud and imaginary crimes.
We are entering a very dangerous world where we will no longer be able to distinguish truth from lies.. That’s fertile ground for autocrats, who benefit when the playing field is muddy, more and more people say that “all politicians are equal,” and the differences between the good and the bad are blurred.
We run the risk of stopping defending principles such as democracy or human rights, because we will have lost the notion of the truth.
What needs to be done? As individuals, we have to verify sources and stop forwarding news and videos that do not come from a credible outlet. As nations, we must support attempts to regulate technology companies, in cooperation with them, to stem the avalanche of falsehoods.
Otherwise, social polarization will be exacerbated, there will be more violence, and our democracies will be further undermined.
The post Why Artificial Intelligence lies to us more and more appeared first on Veritas News.
