OPENAI says that parents will soon have more surveillance on what their teenagers do on Chatgpt.
In a blog post published on Tuesday, the artificial intelligence company extended on its plans has interpreted earlier and in a wider range of situations when it detects potential mental health crises of users who can cause damage.
The announcement of the company comes a week after Openai was struck by its first unjustified death trial, a pair of parents in California who claim that Chatgpt is at fault for the suicide of their 16 -year -old son.
OPENAI did not mention the teenager, Adam Raine, in his Tuesday post. However, after filing the trial, the company alluded that changes were on the horizon.
Over the next month, parents will be able to exercise more control over use by chatgpt adolescents, Openai said. The company will allow parents to connect their accounts with their children, set up rules adapted to age for Chatgpt responses and manage features such as Bot’s memory and cat history.
Parents will soon also be able to receive notifications when Chatgpt detects that their adolescent is “in a moment of acute distress”, according to the Openai blog. This would be the first feature that invites Chatgpt to report conversations from a minor to an adult, a measure that some parents asked for the concern that the chatbot is not able to descend the moments of crisis in itself.
When Adam Raine spoke to GPT-4O of his suicidal ideas earlier this year, the bot sometimes actively discouraged him to seek a human bond, proposed to help him write a suicide note and even advised him on his configuration of a cast node, according to his family’s trial. Chatgpt invited Adam several times with the Hotline de Suicide number, but his parents say that these warnings were easy to get around their son.
In a previous blog article after news from the unjustified Death trial of Raine, Openai noted that its existing guarantees were designed so that the cat to give empathetic responses and returns users to real resources. In some cases, conversations can be sent to human examiners if Chatgpt detects plans to cause physical damage to themselves or to others.
The company said it was planning to strengthen guarantees in longer conversations, where railing is historically more likely to decompose.
“For example, Chatgpt can correctly indicate a hotline of suicide when someone mentions the intention for the first time, but after many messages over a long period, he could possibly offer an answer that goes against our guarantees,” he wrote. “We strengthen these attenuations so that they remain reliable in long conversations, and we are looking for ways to ensure robust behavior on several conversations.”
These measures will be added to the mental health railings that Optai introduced last month, after recognizing that the GPT-4O “failed to recognize the signs of illusion or emotional dependence”. The deployment of GPT-5 in August also came with new security constraints intended to prevent Chatgpt from involuntarily giving harmful responses.
In response to the announcement of OpenAI, Jay Edelson, principal lawyer for the Raine family, said that the CEO of Openai, Sam Altman, should be unequivocal that he thinks that Chatgpt is sure or immediately withdraw from the market. “”
The company has chosen to make “waving promises” rather than withdrawing the product offline as an emergency action, said Edelson in a press release.
“Do not believe it: it is nothing more than the management team of the Openai crisis that tries to change the subject,” he said.
The series of updates focused on security occurs while Openai faces an in -depth examination for the relations of illusions propelles by the AI of people who counted strongly on Chatgpt for emotional support and life advice. Openai had trouble braking excessive chatgpt people, especially since some users got stuck online after the company tried to make the GPT-5 less sycophantic.
Altman has recognized that people seem to have developed a “different and stronger” attachment to AI bots compared to previous technologies.
“I can imagine a future where many people really trust Chatgpt advice for their most important decisions,” wrote Altman in a post last month. “Although it may be great, it makes me uncomfortable. But I expect it to some extent, and soon billions of people could speak to an AI in this way.”
Over the next 120 days, Chatgpt will begin to transport certain sensitive conversations, such as those with signs of a user’s “acute distress”, to Openai reasoning models, who spend more time thinking and working in context before responding.
Internal tests have shown that these models of reasoning follow the security guidelines in a more coherent way, according to the Openai blog.
The company said that it is based on its “expert advice on well-being” to help measure the well-being of users, to set priorities and to design future guarantees. The advisory group, according to Openai, includes experts through the development of young people, mental health and human-computer interaction.
“While the Council will advise you on our product, research and policy products, Openai remains responsible for the choices we make,” wrote the company in its blog article.
The Council will work alongside the “global network of doctors” in Openai, a pool of more than 250 doctors whose expertise that the company claims to be based to enlighten its research in terms of security, its model training and other interventions.
