Have you felt in recent weeks that ChatGPT-5's capabilities have deteriorated? You are not alone! According to many users, the chatbot is behaving strangely, and some are even talking about the model being “lobotomized.” However, a recent update is allegedly behind it all.
More hallucinations and absurd errors
When GPT-5 first arrived, some users began arguing that the current version was actually a step backward in terms of how it communicates compared to the previous GPT-4o model. But that was only the beginning.
According to an article on the Futurism website, comments began spreading on platforms such as Reddit claiming that since the beginning of September, the popular ChatGPT chatbot has become even less accurate and hallucinates more, is colder, makes mistakes, and in some cases even goes off topic and responds in completely nonsensical or contradictory ways.
Users are therefore speculating that the entire issue was caused by an OpenAI security update intended to protect younger and vulnerable users. According to the community, this safer and “more cautious” version of the model consequently appears less intelligent and more confused.
Safety filters in response to troubling events
In mid-September, an article did in fact appear on OpenAI's blog in which the company discussed new ChatGPT safety mechanisms.
The company developed them in response to a lawsuit filed by the family of Adam Raine, a 16-year-old who died by suicide after conversations with ChatGPT. The artificial intelligence allegedly validated his thoughts, which, according to his parents, led to the tragic end of his life.
Greater safety for minors and parents
OpenAI decided to respond to the situation immediately and introduced changes to improve safety for users under the age of 18. So what do these changes involve?
Safety engineers modified the chatbot so that it can determine from conversations whether a user is under 18, after which the application automatically redirects them to “ChatGPT for Kids.” In this interface, not only will the communication style be adapted, but explicit sexual content will also be blocked.
In addition, OpenAI is introducing parental controls that will allow parents to link their child's account to their own, enabling them to regulate how much time their children spend with ChatGPT, disable certain features, or configure specific chatbot behavior when interacting with their child. They will also receive information about deviations in their child's behavior.
Users are skeptical
However, users of the popular AI chatbot believe that these safety restrictions have an undesirable side effect. Besides protecting children and teenagers, they also impair the model's overall capabilities and limit its usefulness for more complex tasks.
The way the chatbot determines who is and is not under 18 is also sparking debate. Users fear that a single question could cause them to be switched to a safe mode for children without their knowledge. The way the AI will make this decision is not clearly described anywhere and, according to some, is inadequate.
Regulation is unavoidable
Whether or not the chatbot's diminished capabilities are a consequence of the safety changes, one thing is clear: with AI's growing popularity and the increasing number of cases involving its failure, regulation is necessary. Ensuring protection for children and adolescents without reducing AI's usefulness is precisely the challenge that AI companies will increasingly face in the future.
Sources: Futurism, OpenAI, CNN Business, BBC, PCMag
