Artificial intelligence (AI) policymaking conversations are taking on a new tenor following the launch (and recall) of Anthropic’s Fable 5 and Mythos 5 models. Fable is a safeguarded version of Mythos, a model Anthropic refused to release to the public in part because of its expertise in detecting cybersecurity vulnerabilities. But in less than a week, the Trump administration moved to ban noncitizen access to both models, to which Anthropic responded by cutting all customer access to Fable and Mythos.

This incident is one manifestation of the larger struggle facing institutions, including school districts: how to respond to AI capabilities that advance faster than institutions can reasonably respond. And while access to Fable/Mythos is gone for now, at some point folks will regain access to them — or to similarly capable models. K-12 leaders and policymakers must move with more urgency to help schools and districts adapt to an era of rapidly improving AI because 1) no one else is going to set that direction and 2) time is not an ally.

Neither of the two large frontier labs, Anthropic and OpenAI, is presenting a compelling vision for how schools and districts should adapt to an age of increasingly capable AI models. Anthropic recently released an Advanced AI Framework and an Economic Policy Framework. The only places schools appear in either of those documents are in lists of critical infrastructure, alongside hospitals and utilities, that need support to better withstand biological or cybersecurity events. OpenAI’s public policy agenda includes a little more on schools, but is limited to calls for work on AI literacy, AI access, and research to understand AI’s effect on schools. Right now, neither of these labs is providing guidance on how schools should adapt to AI as institutions — nor should we expect them to.

Although the education sector is articulating forward-looking visions for K-12 education in an age of AI, more urgency and specificity is needed to translate those visions into frameworks for how advanced AI in how schools do (or don’t) adapt as institutions. The Center on Reinventing Public Education (CRPE) has been doing great work convening leaders to push the K-12 sector to lead rather than react to AI change, but one of their recent publications names a key problem that persists: “The field is over-tooled and under-visioned.” Absent alignment on a vision of schooling in a world with increasingly capable AI, K-12 will be even slower to build concrete plans to drive policy, governance, and practice forward.

Part of this is a familiarity challenge: students aren’t the only ones who need AI literacy education — education practitioners and policymakers need it, too. Many adults in the K-12 sector still have a vision of AI as it existed in a free version of a tool they tried more than a year ago, giving them a skewed sense of where AI capabilities are right now and very little idea of where things are heading in the next few years.

There’s also a challenge of prioritization for leaders in the K-12 sector. Academic performance in many states and districts is still down over the past decade. Enrollment and birthrate declines are placing increasing pressure on budgets and forcing conversations about school closures. And following the successful passage of school cellphone bans, the anti-ed tech wave shows no signs of cresting. In that context, asking schools to adapt to yet another slew of technological shifts seems naive and bordering on tone-deaf.

Mobilizing action to address a force we can’t see or fully understand — a “known unknown” — isn’t an easy proposition. Some might be willing to bet that advanced AI models won’t have a meaningful effect on K-12 schools. If that’s the bet our sector takes either actively or passively and it turns out to be wrong, it’s students who will be affected the most.

Education leaders need to move now to make K-12 institutions more adaptable to advanced AI. CRPE’s recent recommendations offer a solid starting point. Others may choose a different approach, but here are three pillars that should shape leaders’ actions:

• Tracking frontier AI model capabilities is essential: Leading AI tools are evolving quickly in both what they can do and how people interact with them. It’s really hard to adapt to something you don’t understand — staying up to date on AI developments is table stakes for education policymakers and school system leaders.

• Treat AI as one part of a larger vision for change: Schools are already under pressure to improve core academic outcomes, adjust to lower student enrollment, and respond to increased skepticism of how technology is used in K-12. Responses to those pressures should proactively account for advanced AI rather than retrofitting to it later.

• Assume students already have access to frontier AI models: The fact is that no matter what policies schools adopt, students will use — and already are using — using AI outside school. Recent studies show how students’ use of general-purpose AI models is exacerbating longer-term trends of less student cognitive engagement. Adapting to increasingly advanced AI needs to account for this reality, particularly in the design of student assignments and assessments.

The question now, as was true in Issue 10 of this newsletter, is whether schools will lead themselves through this transition or be dragged through it.

Quick Hits

• File under skating to where the puck is moving: A year after the American Federation of Teachers partnered with OpenAI, Anthropic, and Microsoft to train teachers on AI, President Randi Weingarten outlined a 10-point plan that includes pulling screens from pre-K to Grade 2, banning student-facing AI in elementary school, and taxing tech companies to fund AI’s “disruptive” fallout. The erosion of the détente between big tech and teachers unions will add additional fuel to debates over screen time and AI limits during the next round of legislative sessions.
• Canaries aren’t just for coal mines: The Stanford Digital Economy Lab launched The AI Economic Indicators project, which includes several dashboards of data to track AI’s effect on the economy. The “Canaries Dashboard” (a name that’s a little too on the nose) is a productized version of a report we highlighted in Issue 13, showing that early career employment in the most AI-exposed sectors is … not great. As the labor market shifts in response to AI advances, so too will career-connected learning opportunities in K-12 settings.
• In a world where the Knicks are champions, anything is possible: Bellwether co-founder and senior partner Andy Rotherham argues that while project-based learning (PBL) lacks rigorous, equitable, and scalable accountability mechanisms, it’s an area where AI could shift how we think about what’s possible in education. Using AI to improve the rigor and quality of PBL may not survive contact with actual classrooms, but this is one of several opportunities where AI might make old ideas finally work.
• Prompt For America: Anthropic announced Claude Corps, which will place 1,000 fellows in nonprofit organizations with two goals: “that host organizations are equipped with valuable tools and systems, and fellows build AI skills.” None of the announced hosts are K-12 school operators — it’s worth watching if that changes and, if so, what results those partnerships produce.

Odds & Ends

• 🎙️ In case you want more Bellwether perspectives on AI: Mary Wells and Marisa Mission joined Andy Rotherham and Jed Wallace on the WonkyFolk podcast for a wide-ranging conversation on AI and education.

• 📖 From the Bellwether book exchange to your summer reading list: “I Am Not A Robot: My Year Using AI to Do (Almost) Everything” follows Joanna Stern’s quest to integrate AI into nearly every aspect of her life. It’s funny, fast, and honest about both what delivered and what unnerved her during this journey.