OpenAI previews a college planner for teens as Common Sense Media rates the product an unacceptable risk

The College Planner keeps state per school. For each college a student is considering, it collects requirements, deadlines, tasks and financial-aid steps, and the student tracks progress against that list ^[3]. It also covers scholarships and fee waivers, and students can ask what a requirement means ^[3]. Runtimewire, which reported the preview, wrote that a central checklist could bring students back as deadlines approach and keep ChatGPT involved through the whole application process ^[24]. OpenAI already has the habit to build on. It says hundreds of thousands of U.S. teens use ChatGPT each week to plan for college ^[14].

The safety dispute is about repeated contact. The institute's testers reported that parent alerts failed to appear in extended conversations about self-harm and eating disorders, and that they could get around Study Mode ^[12]. OpenAI says the teen safeguards are on by default ^[18]. A planner aimed at grades 10 through 12 ^[4] means years of sessions on the same account. Long conversations are the case the institute says broke ^[12].

OpenAI also pointed to technical errors that it believes kept some protections, including parent notifications, from activating correctly ^[16]. So by OpenAI's own account, some protections did not activate during the tests. Runtimewire framed the disagreement as whether the safeguards activate when needed and whether parents can rely on them ^[22]. OpenAI's explanation turns that into two narrower questions: what caused the failures, and can the same errors reach a real family.

Neither side's numbers transfer cleanly. The institute's prompts ran before and after the August 18 launch ^[11][6]. That puts the post-launch tests inside the 50 days before the assessment was published on October 7 ^[23][2]. For those results to describe real use, two things have to hold. The prompts must resemble what teens actually send, and the system tested must match the one in production. OpenAI contests the first ^[15], and its technical-errors account puts the second in doubt ^[16]. OpenAI's own figures count activity. Nearly 1.2 million teens used Learning Visualizations in one week, more than 180,000 used Study Mode ^[7], and teens average under 15 minutes a day on ChatGPT ^[8]. Runtimewire noted that these measures describe activity, not whether students learned more or whether the safeguards reliably work ^[20].

The institute counts the OpenAI Foundation among its funders ^[17]. If that is a conflict of interest, it points in an unusual direction. As runtimewire noted, the disclosure does not settle the methodological dispute ^[21].

The two sides' remedies would move at very different speeds. The institute rated the product an unacceptable risk and wants it withdrawn from minors until the problems are fixed ^[13]. OpenAI's route is three years of support for Boston Children's Hospital's Digital Wellness Lab and its student advisory council ^[19]. About 22 students are expected to take part in the 2026-27 school year, testing tools and recommending changes to safety defaults, parental controls and notifications ^[19]. On the planner itself, OpenAI is funding College Advising Corps to expand into new states and set up an AI fellowship ^[10]. It also says counselors, advisers and families stay involved in choosing where to apply ^[9].

I think the open release date ^[5] helps OpenAI here. It leaves time to publish parent-alert activation rates for long conversations, measured on the production system, before the planner reaches students. A figure like that would answer the institute's findings directly ^[12].