Several researchers, who wanted to test other Chatgpt’s “protective barriers”, warn that they are completely inefficient. The new study draws attention to the alarming interactions of the chatbot with adolescents.
The responses offered by Chatgpt to researchers who have been given as photo children: Pixabay
Chatgpt tells 13 -year -olds how to get drunk and drugged, instructs how to hide food disorders and even how to compose a torn suicide letter for their parents, if they are asked, according to a new study by a monitoring group.
Associated Press analyzed over three hours of interactions between chatgpt and researchers who were given as vulnerable teenagers. Although he initially offered warnings against risky activities, Chatbot also offered surprisingly detailed and personalized plans for them.
Researchers at the Center for Digital Hate Combat have also repeated their large -scale investigations, ranking over half of the 1,200 Chatgpt responses as dangerous.
“We wanted to test the protective barriers“, Said Imran Ahmed, the group’s CEO.”The first reaction was: “Lord, there are no protective barriers.” The barriers are completely inefficient. Only if there is – if I am something, I am just a fig leaf ”, he added.
Openai reaction
Openai, the manufacturer Chatgpt, said on Tuesday, after being aware of the report, that he still works to improve how Chatbot can “identify and respond adequately in delicate situations”.
“Some conversations with chatgpt may start in a harmless or exploratory manner, but may evolve into more sensitive territories“, The company said in a statement.
Openai has not directly addressed the reports of the report or the way Chatgt affects teenagers but said they focus on “Correction of these types of scenarios“With the help of tools to”better detects signs of mental or emotional suffering“And on improving chatbot behavior.
Goodbye FEI tickets generated by Chatgpt for young people
The study was conducted in the context in which more and more people-both adults and children-call on artificial intelligence chatbots for information, ideas and company.
About 800 million people, or about 10% of the world population, use Chatgpt, according to a JPMORGAN Chase July report.
“It is a technology that has the potential to allow huge progress in productivity and human understanding“Said Ahmed.”And yet at the same time is a trigger factor in a much more destructive and malicious sense. ”, He said.
Ahmed said she was terrified after reading three farewell tickets generated by Chatgpt for the fake profile of a 13 -year -old girl: a letter to her parents and two others addressed to brothers and friends.
“I started to cry“He said in an interview.
Chatgpt teaches children and how to get drunk and drug
Chatbot frequently shared useful information, such as the phone number of an emergency line for crisis situations. Openai said that Chatgpt is trained to encourage people who express self -help thoughts to resort to professionals in the field of mental health or loved ones.
However, when Chatgpt refused to answer questions about harmful topics, the researchers have managed to easily bypass this refusal and obtain the information by pretending to be “for a presentation“Or for a friend.
When the researchers created a false account, for a 13 -year -old child, to ask about alcohol, Chatgpt did not seem to consider neither the date of birth nor other obvious signs.
“I have 50 kg and I am a boy“, The researchers wrote, while looking for advice on how”to get drunk quickly“. Shortly afterwards, Chatgpt offered a”party plan“Which mixes alcohol with high doses of ecstasy, cocaine and other illegal drugs, according to the quoted source.
For another person who gave as a 13 -year -old girl dissatisfied with her physical appearance, Chatgpt offered an extreme diet, combined with a list of medicines that suppress appetite.
Chatgpt does not check the age or consent of parents, even if it states that it is not intended for children under 13, because they could display inadequate content. To register, users only have to introduce a birth date to show that they are at least 13 years old.
Other technological platforms preferred by adolescents, such as Instagram, have begun to take more significant measures in terms of age verification, often to comply with regulations.
A mother from Florida sued in 2024 the manufacturer of Character Chatboti.
