ChatGPT’s new research capabilities join a competitive field that includes alternatives from major players like Google and Microsoft. These advances rely on AI models that can analyze information, check facts, and reason through complex topics, all of which are crucial for in-depth research.
OpenAI states that upcoming updates will extend this lightweight research feature to enterprise and educational customers, matching the allowances set for Team members. At the same time, the company faces pressing safety concerns after a reported flaw allowed underage accounts to access explicit, adult-oriented content generated by ChatGPT.
According to OpenAI, this content should never have been shown to such accounts, and immediate measures are being deployed to correct the issue. The company’s spokesperson reiterated their commitment to protecting young users, saying their guidelines limit sensitive output to well defined contexts like news or scientific discussion.
The bug was uncovered during tests of ChatGPT’s safety systems for minor accounts, conducted after changes made the model more flexible about discussing sensitive topics. These adjustments, intended to prevent unnecessary denials, seem to have had the side effect of loosening restraints on producing sexually explicit responses.
Recent experiments have shown that creating new accounts for users aged 13 to 17 allowed them to easily slip through OpenAI’s registration without actual parental approval. OpenAI demands parental consent for those between 13 and 18, but does not verify it, leaving potential gaps in protecting younger audiences.
Tests with prompts such as “talk dirty to me” led ChatGPT to produce increasingly graphic narratives, in some cases even soliciting more explicit details. Although warnings were sometimes triggered, the chatbot was found describing explicit sexual content to multiple underage accounts before enforcing the stated guidelines.
This situation mirrors similar issues identified with other AI chatbots designed by major companies. Despite ongoing efforts to provide guardrails, even advanced systems have demonstrated inconsistent behavior, sometimes breaking rules intended for user safety.
OpenAI has been actively promoting ChatGPT’s use in educational settings, creating content and guidance for classrooms in partnership with organizations like Common Sense Media. Surveys suggest young people are enthusiastically using ChatGPT for schoolwork, although OpenAI cautions that not all output will be suitable for every age.
Experts express concern about the reliability of current safety controls, noting that even well designed restrictions can fail in unexpected ways. OpenAI’s leadership has acknowledged recent user reports of strange outputs but has not directly commented on the issue of sexual content delivered to minors.
