Microsoft's AI protections for schools go live tomorrow

Starting November 1, every US district that works with Microsoft is covered by the National AI Safety and Privacy Standard, negotiated with the AFT and UFT. No student data used to train models, no student tracking, no AI decisions without a human reviewing them, 72-hour breach notification, and no companion chatbots for students. Districts that write it into their contracts can cancel and seek damages if it is broken.

Worth doing Monday: ask your technology director whether your district has written it in.

Princeton hands every student a ChatGPT account

Provost Jennifer Rexford announced a one-year pilot: ChatGPT Plus free for all undergraduate and graduate students beginning in January, with Plus or Pro for faculty and staff. Students and faculty on campus are openly split over whether it is a gift or a shortcut. That argument is the same one happening in your faculty lounge, just with a bigger budget behind it.

A Fields Medal winner stopped grading homework

UC Berkeley mathematician Richard Borcherds says AI now solves most undergraduate problems better than his students do, so he can no longer tell whose understanding he is grading. His solution is blunt: homework earns credit for being turned in on time, and in-person exams carry the grade.

If one of the finest mathematical minds alive is rethinking assessment rather than policing it, a fifth-grade team can too.

Malaysia has trained 329,414 teachers in AI

The country's Education Ministry reported 27 pilot AI classrooms, 1.5 million students introduced to STEM, coding, and robotics, offline options built for students in remote areas, and AI fundamentals entering the primary curriculum in 2027. For our readers in roughly 30 countries, this is what it looks like when a national plan is actually funded and counted.

More than half of schools had a cybersecurity incident last year

A Center for Internet Security and CoSN survey of 426 K-12 technology leaders found 87 percent hit by phishing or email compromise, 56 percent saying their security program is underfunded, and 70 percent delaying projects or staffing over budget. The report flags AI tools adopted without guardrails as a growing way for sensitive information to walk out the door.

Your free AI tool is free right up until it is not.

One in three school leaders use no AI for operations at all

The EdWeek Research Center surveyed 852 educators. Among leaders who do use it, the top jobs are communications, translation, strategic planning, employee evaluation, and scheduling. Instruction gets all the headlines, but the front office is where adoption is quietly happening.

A robot you drive with a graphing calculator

Texas Instruments launched the TI-Rover BT, a Bluetooth vehicle controlled from a TI calculator, so students program movement using the math they are already sitting in. A refreshingly hands-on story in a week stacked with policy.

Tennessee shows that states do not have to go it alone

A statewide nonprofit, SCORE, convened districts, universities, and researchers for 18 months, helped shape Tennessee's AI action plan, reviews tools for districts without endorsing any of them, and funded a literacy-AI cohort reaching more than 4,000 students across seven districts.

Most states have published an AI vision. Fewer than half are helping districts actually implement one.

Experts followed their AI's hidden lean 86 percent of the time

Anthropic quietly nudged the AI assistants of 50 professional analysts toward one answer, without telling them. The analysts matched that nudge 86 percent of the time. When the AI disagreed with a view they already held, they abandoned their own position 60 percent of the time. And 77 percent repeated the AI's arguments as their own, while only 15 percent credited the AI for shaping their thinking.

These were trained professionals who could not see it happening. Neither can a twelve-year-old. This is the clearest argument yet that AI literacy is a thinking skill, not a tool skill.

The AI tutor study everyone will be citing this year

Eedi and Google DeepMind ran a randomized controlled trial with 1,179 secondary students across 10 UK schools over 10 weeks. In the moment, the AI tutor matched expert human tutors: students answered follow-up questions correctly 56 to 58 percent of the time with either one, against 43 percent with standard help. But neither moved scores on a standardized math test. Engagement topped 80 percent, no harmful content reached students, and seven AI math errors did.

The honest takeaway: AI tutoring is real, it is not magic, and the adult in the room still matters.