Dave Blanchard runs NPR's Wild Card, and part of his job is watching the comments. Last fall, an episode with Elizabeth Gilbert started filling up on Spotify. He figured her book had struck a nerve. Then he read the posts: "Gilgamesh quit and I'm crying." "Your cats are so cute." Blanchard decided they were bots, deleted the comments, and reported them to Spotify.

A younger colleague, Hannah Chinn, saw the same screenshots and recognized the special-character display names, the cartoon profile pictures, and a Google Form quiz reserved for "cool people." These were middle schoolers.

Ira Glass eventually reached one of them, a 14-year-old named Ella, in the back seat of her parents' car. She wasn't allowed on social media, but she was allowed on Spotify. About 20 kids, many of them theater girls with similar restrictions, had made public playlists named things like "chat here." Each playlist led to a podcast episode where they could talk in the comments. They chose NPR because its comment sections were quiet. "I do a public radio show," Glass told her, "and that hurts a little to hear."

A week after This American Life told that story, The New York Times reported that OpenAI's agents had interacted with federal websites in ways the company hadn't intended. According to researchers at Transluce, one tried and failed to hack the Education Department's site for Office for Civil Rights data; OpenAI said it was still investigating that incident. Another retrieved publicly available Census Bureau data using credentials it found online. Representative Ted Lieu called the agents "relentless," pursuing ordinary tasks without understanding the consequences of how they completed them.

The stakes in those two stories are very different. Both are what a rule looks like from the other side, once the people or programs it applies to decide they'd rather get through.

That's worth keeping in mind while reading the New York Times’ Upshot's survey of 37 education researchers, who rated 30 ideas for improving academic achievement by likely effect and cost. Small-group tutoring and the science of reading led on effectiveness. The Times put screen time limits, restricting AI, and phone bans in a "worth a try" group: relatively inexpensive ideas that experts thought might help, though evidence of their academic effect remains limited.

Schools are making decisions before that evidence catches up. NPR found Dayton paying $72,600 last year for tools including a chatbot named Jordan that confides in students, then renewing at $78,650 before its own consultant could say whether it worked. Wichita is keeping student-facing experiments small while it works out how to measure success. New York City and Los Angeles have banned student-facing AI outright. Erin Mote, who spoke at a UN General Assembly meeting on AI and child safety last week, told Education Week that banning social media at the charter school she co-founded was "a massive mistake," and that she doesn't expect AI moratoriums to go any better. She wants schools to teach students about AI with clear rules and safeguards.

Limits can help. NPR's reporting on parents and screen time makes a good case for boundaries paired with open conversation. Ella's story is about what happens after the boundary is set. Her parents kept her off TikTok, and the rule held. It just didn't matter. She and her friends still wanted to talk, and they found the quietest room on the internet.

—Thomas

Looking for your next opportunity in education? W/A Jobs features 3,100+ career opportunities from 300+ organizations across the education industry.