What happened in the room
On October 5 in Washington, OpenAI convened a forum on teen AI safety, moderated by Chris Lehane, the company's chief global affairs officer. The panel brought together Arthur C. Evans Jr., chief executive of the American Psychological Association, Kate Ruane, who directs the Free Expression Project at the Center for Democracy and Technology, Mathilde Cerioli, chief scientist at the youth-AI nonprofit everyoneAI, and Chuck Melley, senior vice president of policy, advocacy, and communities at CodeAI. The composition, clinicians, civil-liberties advocates, and AI-literacy practitioners in the same room, signaled OpenAI wanted a harder conversation than a product showcase.
Evans set the tone early: "We can't sit back and wait." That urgency ran through the panel, but the specifics stayed close to design philosophy rather than policy mechanics. Melley argued the standard for a useful tool is narrow: AI "can help them deepen their thinking, as long as it doesn't give them the answer." Cerioli raised the harder design question, that a tool a student depends on to produce an answer is only useful until it's removed, at which point "they don't know how to do it." That framing, dependency versus skill-building, is the actual design problem any ed-tech buyer should be asking vendors about.
The proposals on the table
The ideas panelists floated were concrete enough to write down, if not yet to enforce. Ruane pushed for AI literacy standards written into curricula, framed as a direct question: "How do we actually teach kids how to become experts while using this technology?" Evans and Melley both backed funded professional development, arguing that teachers cannot model safe AI use for students if districts never budgeted time or money to train them on it first. Cerioli's recommendation was more technical: design tools around adolescent developmental stages specifically, rather than shipping the same assistant built for a 35-year-old knowledge worker into a tenth-grade classroom.
The panel's final and more structural idea was to shift the market's incentive away from restriction and toward competition, encouraging AI companies to compete on who builds the safest product for teenagers rather than waiting for regulators to set a floor first. That is a reasonable idea in theory. It is also the same bet the industry has made before, shipping a consumer AI product into classrooms and patching the safety gaps only after districts and parents noticed the damage. Nothing discussed at this forum changes that sequence. It simply names the pattern more explicitly than OpenAI has before.
Why this forum happened now
The timing traces directly to two events earlier in the fall. New York City and Los Angeles, two of the largest school systems in the country, both moved to restrict AI tools in classrooms this year after losing confidence that vendors would self-regulate fast enough. Weeks later, Microsoft signed a legally binding AI guardrails agreement with the American Federation of Teachers, turning voluntary promises into audited contract terms for every one of its school accounts starting November 1. Those two moves put real pressure on OpenAI, whose products sit in far more classrooms informally than any binding agreement currently covers.
A voluntary forum is a far cheaper response than a binding contract, and the gap between the two is the real story here. OpenAI convening psychologists and advocates to discuss teen safety is a genuinely useful exercise, and it is also the kind of move a company makes when it wants to be seen responding to a moment before accepting the enforcement terms that moment is starting to demand from its competitors. Nothing said on the panel commits OpenAI to an audit, a contract clause, or a deadline the way Microsoft's deal does. That distinction will matter more once districts start asking vendors to show their homework.
The gap between a forum and a contract
Compare what came out of this forum with what Microsoft already signed. Microsoft's deal bars training on student data except for narrow, defined exceptions, bans engagement-maximizing companion features outright, and requires independent third-party audits starting on a fixed date. OpenAI's forum produced four directional ideas, literacy standards, funded teacher training, developmentally aware design, and a plea for safety competition, none of which carry a deadline, an audit, or a penalty for non-compliance. One is a procurement document. The other is a position paper.
That gap is not a knock on the people in the room, who raised real and specific problems worth solving. It reflects something simpler: a forum sits upstream of policy, while the companies actually setting enforceable terms right now are negotiating directly with the institutions that buy their products at scale. CIOs should read OpenAI's forum as a signal of where standards are heading, not as evidence that OpenAI has already met them.
What this tells us about where regulation is heading
Put side by side, these two events describe a fairly predictable sequence: informal use expands fastest, controversy forces union or district action, one major vendor negotiates a binding standard to get ahead of legislation, and competitors respond with forums, frameworks, and literacy initiatives that read as responsible while carrying none of the same enforcement weight yet. Illinois already shows the next stage: state legislatures translating pieces of this fight directly into statute, in that case banning AI from teacher evaluations outright starting January 2027.
Expect more states to follow with narrower, more specific rules rather than comprehensive AI-in-schools legislation, because narrow rules are easier to pass and harder for a vendor to contest. CIOs managing any AI deployment that touches minors, from K-12 tools to a corporate learning platform used by interns, should assume today's voluntary commitment is next year's statutory floor, and negotiate contract language now that would survive if a regulator wrote exactly these forum proposals into law.
What to do with this before it becomes a mandate
Treat OpenAI's forum as a checklist of questions to ask any AI vendor serving a younger or vulnerable population today, not next year. Ask specifically how the product is designed around developmental stage rather than adult defaults, what teacher training the vendor funds or subsidizes, and what AI literacy curriculum, if any, ships alongside the tool. None of those questions require a new law to ask; they require a procurement process willing to hold a vendor to the standard its own executives floated in public.
The companies that can already answer those questions with specifics, the way Microsoft now can because its answers are written into a contract, will win public-sector and youth-facing deals over the next two budget cycles. The vendors still answering with a forum transcript instead of an audit clause are telling you something true about where they actually stand, whatever the panel said in the room. Build that distinction into your next vendor scorecard.
