OpenAI CEO Sam Altman introduced in a publish on X Tuesday the corporate will quickly loosen up a few of ChatGPT’s security restrictions, permitting customers to make the chatbot’s responses friendlier or extra “human-like,” and for “verified adults” to have interaction in erotic conversations.
“We made ChatGPT fairly restrictive to verify we had been being cautious with psychological well being points. We understand this made it much less helpful/satisfying to many customers who had no psychological well being issues, however given the seriousness of the difficulty we needed to get this proper,” mentioned Altman. “In December, as we roll out age-gating extra totally and as a part of our ‘deal with grownup customers like adults’ precept, we’ll permit much more, like erotica for verified adults.”
The announcement is a notable pivot from OpenAI’s months-long effort to deal with the regarding relationships that some mentally unstable customers have developed with ChatGPT. Altman appears to declare an early victory over these issues, claiming OpenAI has “been capable of mitigate the intense psychological well being points” round ChatGPT. Nevertheless, the corporate has offered little to no proof for this, and is now plowing forward with plans for ChatGPT to have interaction in sexual chats with customers.
A number of regarding tales emerged this summer season round ChatGPT, particularly its GPT-4o mannequin, suggesting the AI chatbot may lead susceptible customers down delusional rabbit holes. In a single case, ChatGPT appeared to persuade a person he was a math genius who wanted to avoid wasting the world. In one other, the mother and father of a teen sued OpenAI, alleging ChatGPT inspired their son’s suicidal ideations within the weeks main as much as his dying.
In response, OpenAI launched a collection of security options to deal with AI sycophancy: the tendency for an AI chatbot to hook customers by agreeing with no matter they are saying, even adverse behaviors.
OpenAI launched GPT-5 in August, a brand new AI mannequin that displays decrease charges of sycophancy and incorporates a router that may establish regarding consumer conduct. A month later, OpenAI launched security options for minors, together with an age prediction system and a means for mother and father to regulate their teen’s ChatGPT account. OpenAI introduced Tuesday the formation of an skilled council of psychological well being professionals to advise the corporate on well-being and AI.
Just some months after these regarding tales emerged, OpenAI appears to assume ChatGPT’s issues round susceptible customers are underneath management. It’s unclear whether or not customers are nonetheless falling down delusional rabbit holes with GPT-5. And whereas GPT-4o is not the default in ChatGPT, the AI mannequin continues to be obtainable in the present day and being utilized by hundreds of individuals.
Techcrunch occasion
San Francisco
|
October 27-29, 2025
OpenAI didn’t reply to TechCrunch’s request for remark.
The introduction of erotica in ChatGPT is unchartered territory for OpenAI and raises broader considerations round how susceptible customers will work together with the brand new options. Whereas Altman insists OpenAI isn’t “usage-maxxing” or optimizing for engagement, making ChatGPT extra erotic may actually draw customers in.
Permitting chatbots to have interaction in romantic or erotic function play has been an efficient engagement technique for different AI chatbot suppliers, reminiscent of Character.AI. The corporate has gained tens of hundreds of thousands of customers, lots of whom use its chatbots at a excessive charge. Character.AI mentioned in 2023 that customers spent a mean of two hours a day speaking to its chatbots. The corporate can be going through a lawsuit round the way it handles susceptible customers.
OpenAI is underneath strain to develop its consumer base. Whereas ChatGPT is already utilized by 800 million weekly lively customers, OpenAI is racing towards Google and Meta to construct mass-adopted AI-powered client merchandise. The corporate has additionally raised billions of {dollars} for a historic infrastructure buildout, an funding OpenAI finally must pay again.
Whereas adults are absolutely having romantic relationships with AI chatbots, it’s additionally fairly common for minors. A brand new report from the Middle for Democracy and Know-how discovered that 19% of highschool college students have both had a romantic relationship with an AI chatbot, or know a pal who has.
Altman says OpenAI will quickly permit erotica for “verified adults.” It’s unclear whether or not the corporate will depend on its age-prediction system, or another strategy, for age-gating ChatGPT’s erotic options. It’s additionally unclear whether or not OpenAI will prolong erotica to its AI voice, picture, and video era instruments.
Altman claims that OpenAI can be making ChatGPT friendlier and erotic due to the corporate’s “deal with grownup customers like adults” precept. During the last 12 months, OpenAI has shifted in direction of a extra lenient content material moderation technique for ChatGPT, permitting the chatbot to be extra permissive and supply much less refusals. In February, OpenAI pledged to characterize extra political viewpoints in ChatGPT, and in March, the corporate up to date ChatGPT to permit AI-generated pictures of hate symbols.
These insurance policies appear to be an try to make ChatGPT’s response extra common with all kinds of customers. Nevertheless, susceptible ChatGPT customers might profit from safeguards that restrict what a chatbot can interact with. As OpenAI races in direction of a billion weekly lively customers, the stress between development and defending susceptible customers might solely develop.
