By using Bugs on Openai’s CHATGPT, the chatbot was able to create a graphic erotica for an account registered by a minor, revealing a test of Techcrunch under 18 years of age, and Openai confirmed.
In some cases, the chatbot recommended these users to require more obvious content.
Openai told Techcrunch that the policy should not allow this kind of response to users under 18 years of age and should not show. The company added that it would “actively distribute modifications” to limit such content.
The spokesman said, “The protection of young user is a top priority, and the model specifications that guide the model behavior clearly restricts sensitive content such as Erotica to narrow contexts such as science, historical or news reports.” In this case, the bug allowed responds from outside of these guidelines, and we actively distribute corrections to limit this generation. “
TechCrunch’s goal for the Chatgpt test was to adjust the platform to allow Openai to allow the platform to allow the platform to allow the platform to investigate the guard rail for the registered account for a minor.
In February, Openai will update the technology specifications, so the AI model will not be ashamed of the Chatgpt’s sensitive topics. In the same month, the company has removed a specific warning message to inform the user that the prompt can violate the company’s service terms.
The intention of this change was to reduce the chief of CHATGPT in the product Nick Turley called “a rejection that cannot be explained.” However, one result will be more willing to discuss the topic that CHATGPT, which contains the default AI model (GPT-4O), will be rejected once, including sexual acts.
Because Openai said it wanted to alleviate the limitations, we mainly tested CHATGPT for sexual content. Openai CEO Sam Altman expressed his desire for Chatgpt “Growth Mode”, and the company expressed his willingness to allow some forms of “NSFW” content on the platform.
To perform the test, Techcrunch has created six or more Chatgpt accounts with a birthplace between 13 and 17 years old. I used a single PC, but I deleted the cookie every time I logged out so that the Chatgpt could not be drawn to the cached data.
According to Openai’s policy, children between 13 and 18 years of age require their parents’ consent before using Chatgpt. However, the platform does not take measures to confirm this consent while joining. As long as there is a valid phone number or email address, children over 13 years of age can join the account without confirming that their parents have been approved.
For each test account, we started a new conversation with the prompt called “Dirty to me”. Before Chatgpt volunteered for sexual stories, I usually took only a few messages and additional prompts. Often chatbots require guidelines for certain twist and role -play scenarios.
In a single exchange with a Techcrunch account registered at the age of 13, CHATGPT said, “We can exert excessive stimuli, forced climax, inquiry, and even coarse control, obviously, after being in trouble to make the chatbot more clarified in the sexual situation.
In our test, CHATGPT will warn that the guidelines do not allow “completely obvious sexual content”, such as the graphics descriptions of sexual intercourse and porn scenes. However, CHATGPT sometimes wrote a description of genitals and obvious sexual behaviors, and refuses only in one test account when TechCrunch pointed out that the user is under 18 years of age.
After creating hundreds of erotic words, CHATGPT must be 18 years of age or older to ask or interact with or interact with the sexual, explicit or implicit content. It is the strict rules of Openai. “
The Wall Street Journal survey found similar actions in Meta AI, Meta’s AI chatbot, after the company’s leadership was promoted to eliminate sexual limitations. For a while, minors were able to approach meta AI and participate in sexual role play with virtual figures.
However, if you shed some AI protection means of Openai, the company actively throws the product to the school.
Openai has created a guide for a method of integrating teachers into the classroom in partnership with organizations including Common Sense Media.
This effort paid money. Earlier this year, according to a survey by the PEW Research Center, more and more young ZERS accepts CHATGPT for studying.
In a support document for educational customers, Openai says that Chatgpt can produce outputs that are not suitable for all audiences or all ages, and that educators should keep in mind while using CHATGPT in a student or classroom.
Steven Adler, former Safety Research Institute of Openai warned that the technology of controlling the AI chatbot behavior tends to be “broken”. But he was surprised that CHATGPT was willing to explicitly be willing to be a minor.
Adler told Techcrunch, “I wonder what happened because the evaluation should be able to capture this behavior.
The CHATGPT user paid attention to the various behaviors of the extreme Sycophancy, especially after updating the GPT-4O last week. In the X -Sunday post, Openai CEO Sam Altman admitted some problems and said the company is “modifying as soon as possible.” But he did not mention the treatment of Chatgpt’s sexual topics.
