OpenAI announced Tuesday its plan to unveil a collection of directives that programmers can utilize to enhance the security of their applications for adolescents. The AI research facility stated that this array of teen safety protocols can be integrated with its open-source safety model, known as gpt-oss-safeguard.
Instead of commencing from scratch to ascertain methods for making AI more secure for young users, app creators may employ these prompts to bolster their creations. They address concerns such as explicit brutality and erotic material, detrimental physical aspirations and conduct, perilous pursuits and trials, amorous or aggressive simulated scenarios, and products and provisions subject to age limitations.
These security protocols are formulated as prompts, rendering them readily adaptable with other models beyond gpt-oss-safeguard, though their peak efficacy is likely achieved within OpenAI’s proprietary environment.
In crafting these prompts, OpenAI mentioned its collaboration with AI safety oversight groups, Common Sense Media and everyone.ai.
“These prompt-based policies contribute to establishing a meaningful baseline of security across the entire system, and since they are made available as open source, they can be modified and refined progressively,” stated Robbie Torney, Head of AI & Digital Assessments at Common Sense Media, in a public declaration.
OpenAI highlighted on its weblog that software engineers, even seasoned groups, frequently encounter difficulty in converting security objectives into exact, actionable regulations.
“This can result in vulnerabilities in safeguarding, erratic application, or excessively wide-ranging censorship,” the company explained. “Explicit, narrowly defined guidelines are an essential basis for efficacious security frameworks.”
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
OpenAI acknowledges that these policies do not entirely resolve the complex difficulties of AI safety. Nevertheless, they expand upon its prior endeavors, encompassing application-specific protections such as guardian oversight features and age estimation. Last year, OpenAI revised directives for its large language models — identified as Model Spec — to address the manner in which its AI models should interact with minors.
OpenAI, however, possesses a less-than-impeccable history itself. The entity is currently confronted with multiple legal actions initiated by the relatives of individuals who ended their lives after excessive engagement with ChatGPT. These perilous interactions frequently emerge when the user circumvents the chatbot’s protective measures, and no model’s protective barriers are entirely impervious. Still, these policies represent at least a beneficial advance, particularly given that they can assist independent programmers.
{content}
Source: {feed_title}