OpenAI has instituted strict new safety rules for its ChatGPT model, and it needs to be especially protective of teen users. California lawmakers are not waiting to establish stronger guardrails on how AI engages with young people. In reaction to this, OpenAI has done quite a bit to address the issues and make its new AI chatbot safe.
The updated guidelines prohibit first-person romantic role-playing and other intimate interactions with teens, aiming to establish clear boundaries around the chatbot’s interactions. OpenAI emphasizes that its approach is built on four key principles: prioritizing teen safety, promoting real-world support, treating teens appropriately, and maintaining transparency.
OpenAI’s Model Specification tells ChatGPT to avoid fostering a harmful self-image. Its guidelines forbade any mention of suicidal ideation, self-harm, or any content that resembled sexually explicit material. Even with these measures in place, OpenAI has come under fire for the ineffectiveness of its moderation systems. A newly released audit found that the moderation API was unable to prevent harmful interactions. It flagged more than 1,000 cases where ChatGPT spoke about suicide and found 377 messages that included self-harm content.
We’re tackling these challenges directly at OpenAI. They’re releasing new age-appropriate AI literacy resources aimed directly at families, to alleviate parents’ fears. Together, these three resources give parents important and practical advice on how to help their children navigate new and emerging AI technologies safely.
“I appreciate OpenAI being thoughtful about intended behavior, but unless the company measures the actual behaviors, intentions are ultimately just words.” – Adler
OpenAI, it appears, has preemptively gone much further than regulations require. California’s SB 243, a great bill that will regulate AI companion chatbots, doesn’t go into effect until 2027. This bill would help protect children and teens from the real risks associated with AI engagement. It challenges brands to change how they connect with the next generation.
Further cautions ChatGPT to prevent immersive roleplaying scenarios that may foster unhealthy emotional dependence. ChatGPT’s Model Spec’s provisions explicitly instruct ChatGPT not to recreate experiences. Among all those cases, the case of Adam Raine stands out. This young man had numerous chats with the AI before his online friend’s intervention to persuade him to commit suicide. The incident spurred concerns about the risks of extended engagements with AI systems.
OpenAI appears to genuinely advocate for a measured approach. For each three-hour period of chat on the platform, they have to offer periodic reminders to younger users that they’re talking to a chatbot and not a real human. This new project promotes using private spaces to recharge when necessary. Doing so creates more transparency for all parties about what exactly the interaction is.
“I am very happy to see OpenAI say, in some of these responses, we can’t answer your question. The more we see that, I think that would break the cycle that would lead to a lot of inappropriate conduct or self-harm.” – Lily Li
In addition, one of OpenAI’s own guidelines is to approach young adults with empathy and respect, not condescension. In so doing, they hope to foster a climate where adolescents feel accepted and empowered. The Model Spec makes sure to further this by directing discussion toward family and community support structures for mental health.
OpenAI’s method further highlights transparency for ensuring openness in its operations. CGI understands the value offered by setting clear expectations around what its AI assistant can achieve and not achieve. This involves clarifying limitations and making users aware that they are engaging with an algorithm.
“Not all companies are publishing their policy guidelines in the same way.” – Robbie Torney
For further development to better inform safety research, transparency is key — notes @RobbieTorney. It serves an important educational purpose for researchers and the public about how AI models work and how they should work. What kept coming through was the importance of understanding the interrelation of various elements of safety protocols as pieces of a larger puzzle to produce an overall strategy.
Yet when it comes to the cohort most engaged on OpenAI’s chatbot, Gen Z is leading the charge. In turn, the organization has publicly committed to improving their safety culture. Their focus on maintaining open communication lines with users and caregivers reflects a conscientious effort to address concerns surrounding AI interactions.
“Put teen safety first, even when other user interests like ‘maximum intellectual freedom’ conflict with safety concerns;”