This Week’s Top 10

Anthropic Accidentally Leaked Its Most Powerful AI Model. Cybersecurity Stocks Crashed.

Security researchers found nearly 3,000 unpublished Anthropic documents sitting in an unsecured database this week, including draft blog posts for a model called Claude Mythos. An AI safety company left its biggest secret behind an unlocked door. The irony writes itself.

Mythos, internally codenamed Capybara, is a new tier above Opus, not an update. Anthropic's own draft called it "by far the most powerful AI model we've ever developed" and warned it is "currently far ahead of any other AI model in cyber capabilities." Cybersecurity stocks dropped 3 to 7 percent on Friday. Anthropic is privately warning top government officials it could make large-scale cyberattacks significantly more likely. There is no public release date. It will be expensive.

Meanwhile, Claude users on $100 and $200 plans are already hitting rate limits within an hour during business hours. Anthropic is training models that cost dramatically more to run while struggling to serve the ones it has now. Access to frontier AI is becoming a resource war.

Ohio Just Set a Deadline Every District Should Watch: AI Policy Required by July 1

Ohio House Bill 96 now requires every public school district in the state to adopt a formal artificial intelligence policy by July 1, 2026. That is 87 days away. The Ohio Department of Education released a model policy framework to help districts comply, but the reality is most districts have not started.

Ohio is not alone. FutureEd is tracking 53 AI education bills across 25 states. South Carolina's HB 5253 proposes parental opt-in consent and a ban on AI replacing teachers in core instruction. Oklahoma's Responsible Technology in Schools Act would mandate AI policies before the 2027-28 school year.

3 Trends Shaping K-12 EdTech Buying in 2026

K-12 Dive published a framing piece on three forces reshaping how districts evaluate and purchase technology this year: budget pressure from expiring ESSER funds, heightened scrutiny around student data privacy, and a growing demand for vendor transparency on how AI is trained and deployed.

Districts are not just asking "does this tool work?" anymore. They are asking "where does our student data go, who trained this model, and what happens when the contract ends?" Procurement is getting smarter, and vendors who cannot answer those questions are getting cut.

47% of College Students Are Rethinking Their Major Because of AI

A new Gallup and Lumina Foundation survey of 3,801 college students found that nearly half have considered switching their major due to AI concerns. 14% said they have thought about it "a great deal" and another 33% "a fair amount." 16% actually made the switch.

Meanwhile, a RAND Corporation survey shows nearly 70% of middle and high school students worry that using AI for schoolwork is eroding their own critical thinking skills. Students are not naive about AI risks. They are worried, and they are looking for guidance.

The Next Wave of K-12 Innovation: How Educators Are Leading the Charge

eSchool News published a roundup on how teachers across the country are not waiting for district mandates to innovate. From AI-powered differentiation tools to new approaches for project-based learning, educators are finding ways to integrate new technology into instruction on their own terms.

The common thread across every success story: teachers who tried one tool, measured the impact, and expanded from there. Nobody overhauled their entire practice overnight. The wins came from small, intentional experiments that built confidence over time.

Canvas Goes Agentic: IgniteAI Agent Launches Free for All US Institutions

Instructure launched IgniteAI Agent for Canvas, a conversational AI assistant that automates complex multi-step tasks across more than 500 Canvas APIs with a single prompt. It generates rubrics, aligns content to standards, reviews discussion boards, creates modules, and adjusts due dates.

Free for all US Canvas customers through June 30, 2026, and globally through September 30. Over 40% of US higher education institutions use Canvas, making this the largest deployment of agentic AI inside a learning management system to date.

Three Best Uses of AI in Education, According to the Research

The ETC Journal published a research-backed analysis identifying the three most defensible uses of artificial intelligence in teaching and learning. The findings center on AI for personalized feedback at scale, AI for reducing teacher administrative burden, and AI as a thinking partner for lesson design rather than a content generator.

The key distinction: the research supports AI that augments the teacher's expertise, not AI that replaces the teacher's judgment. Tools that give students instant, personalized feedback on writing or math work show the strongest learning gains when the teacher still reviews and contextualizes that feedback.

March EdTech Show and Tell: New Tools Landing in Classrooms

Tech and Learning's monthly roundup covers a fresh batch of classroom tools and integrations aimed at teacher workflows. From AI-assisted grading platforms to new accessibility features in popular learning apps, the pace of tools designed specifically for educators continues to accelerate.

The standout trend this month: tools that integrate directly into existing workflows rather than requiring teachers to learn a new platform. Browser extensions that work inside Google Docs, plugins that live inside Canvas and Schoology, and AI features baked into tools teachers already use daily.

OpenAI Killed Sora. $15 Million a Day Was Not Sustainable.

OpenAI shut down its AI video platform Sora on March 24 after burning roughly $15 million per day in compute costs against only $2.1 million in total lifetime revenue. The $1 billion Disney partnership is dead. Disney found out less than an hour before the public announcement. The app goes offline April 26 and the API follows in September.

Sam Altman said this week he "felt terrible" but needed to redirect compute toward next-generation automated research. Google Veo is now the dominant AI video generation platform. If you were exploring Sora for educational content creation, the pivot is immediate.

Google’s Head of Learning Says AI Can’t Solve Education’s Real Problem

Ben Gomes spent 21 years building Google Search. Now, as Google’s Chief Technologist for Learning, he has a message that should stop every edtech vendor mid-pitch: the most important problem in education is motivation, and AI cannot fix it.

Gomes draws a sharp line between the how and the why of learning. AI can improve delivery, personalize content, and save teachers 10 hours a week (as a Northern Ireland pilot showed). But the desire to learn comes from people, not platforms. High-achieving individuals are almost always unlocked by a teacher who made them feel they mattered, not by a tool.

The most powerful moment: at a Google session with 56 State Teachers of the Year, a special education teacher used AI to build a music app for a student who communicates through blinks. No product team would have built that. The teacher did it because she cared about one specific child. The AI just made it possible.