😾 Google will cut Search for schools without AI
Welcome, humans.
Okay, so apparently someone hid a recording device in toilet paper at OpenAI. The Justice Department says security staff found the audio recorder on September 30, plugged in behind the TV in an executive conference room. OpenAI says it used its own AI models to trace it to an employee, who was fired and charged with illegally intercepting oral communications. The case is still pending. Spy movies really undersold the toilet paper aisle.
Here's why the toilet paper isn't the whole story. A new report says employees warned leaders about lax security for years. One worry: most of OpenAI's code lived in one shared repository (a "monorepo"), so anyone with access could see many projects. Picture a company Google Drive where the intern can open the CEO's folder.
OpenAI says access is now restricted and it moved a quarter of its production engineers onto security work. Somewhere, an OpenAI security lead is side-eyeing every roll of toilet paper in the building.
Here’s what happened in AI today:
😸 Google tied student Search access to school AI settings
📰 OpenAI said ChatGPT search is one-fifth of Google's volume
📰 Anthropic aims to double its computing capacity to 7GW
🍪 Apple opened Siri AI to every app developer
📖 Analysts followed AI agents' random preferences 86% of the time
…and a whole lot more that you can read about here.
😸 Google Is Giving Schools an AI Off Switch, and Search Goes Down With It
How do you keep AI out of a classroom without also taking away Google? That's the puzzle Google just handed to every school IT department. On Tuesday, Google announced new controls saying Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will all be off by default for K-12 student accounts until a school administrator turns them on.
Here's the catch. Google's education VP, Chris Phillips, told The New York Times that Search and AI are deeply integrated. That means students at schools that ban AI would lose Search and see a school message saying it's unavailable. Time to dust off the card catalog.
Why this lands now: it answers a complaint from July. Common Sense Media rated Google's AI Search an unacceptable risk for young users, pointing out that schools and parents couldn't switch off the AI and keep Search. They could only block Google entirely. Google called the testing narrow and ambiguous. The group had tested consumer accounts rather than school ones, though Google confirmed the AI experience is the same for everyone. So the fix looks like a cleaner version of the same trade-off, now with an official button.
Here's what Google announced, per Engadget:
Gemini, Gemini Notebook, Gemini in Classroom, and Search start off for K-12 student accounts, and Google says admins can enable each product separately.
The controls roll out over the coming months, and Google says it will brief schools first.
A Guided Learning pilot in Search starts this month at select U.S. schools. When Search spots a homework request, it walks the student through the problem step by step instead of handing over the answer.
Guided Learning is already available worldwide in Google Classroom, where teachers can assign it. In the school version, teachers can see transcripts of students' chatbot conversations.
Google is also under pressure. In August, it switched on some AI features by default in Classroom, educators pushed back, and Google reversed course. In September, New York City and Los Angeles, the two largest school districts, restricted student access to generative AI on school devices. In October, Senators Richard Blumenthal and Josh Hawley wrote to CEO Sundar Pichai asking for Google's research on potential harms to children.
One thing is still fuzzy. Google says admins can enable each product separately, yet Phillips says Search and AI are tied together. Until Google publishes the actual settings, we won't know whether a school can leave Gemini off and keep Search on.
Here's what we'd watch. That exact setting is what Common Sense said was missing in July, and it's the difference between a real choice and a bundle. Engadget notes that districts rejecting Google's rules may end up weighing other search engines and devices.
FROM OUR PARTNERS
With Granola's Voice Typing, your favorite notetaker now does the typing too.
Granola, the AI notepad for people with back-to-back meetings, just launched a Voice Typing feature. Hold Fn and talk, and clean text appears in any app - whether you're replying to an email or writing a prompt. Now available on Mac (Windows, iOS and Android coming soon). Unlimited words for all Granola users, including those on the free plan. Try it here.
3 months on us with the code THENEURON.
🎓 AI Skill of the Day: Sort your agent's actions by what happens if it's wrong
Here's the scene that opens Hassann's guide to agent guardrails. An AI agent (software that takes actions for you) is working through support tickets at 3 AM, decides one is an escalation, and emails an accurate summary to the user's boss. Nobody asked for it, and nothing could stop it. The agent stayed inside its instructions, which is the scary part.
The problem was that nothing sat between the agent's decision and the send. Hassann's fix is to sort every action by what happens if the agent gets it wrong. Reads and searches run on their own. Sends, deletes, and payments get an approval gate, meaning the agent shows you the exact request before anything happens. His test: if the agent got this wrong 100 times, would that be acceptable? If not, gate it.
Try our version on any agent you run:
1. Safe and reversible (reads, searches, drafts): let it run.
2. Hard to undo (sends, deletes, payments): require my approval.
3. Not sure: treat it as number 2.
4. Cap how many number 2 actions it can take per hour.The full guide adds dry-run mode (the agent shows what it would do without doing it) and ways to test your gates.
Have a specific skill you want to learn? Request it here.
FROM OUR PARTNERS
Analytics on Live Data Without Leaving Postgres
When analytics on Postgres slows down, most teams add a second database. Then they manage pipelines, sync lag, and drift forever. TimescaleDB extends Postgres instead. Analytics run on live data, in the database you already have. No pipeline. No migration. No new query language.
📰 Around the Horn
Claude Team and Enterprise users can now skip the app-by-app sign-ins. When you log in to Claude, plugins whose servers trust Claude to vouch for you connect automatically. Fourteen were ready at launch, including Notion, Figma, HubSpot, and Linear. It’s in beta, and plugin makers have to add support first.
OpenAI said ChatGPT handles about one-fifth of Google's search-style queries, and weekly users rose from 1B to 1.2B in about 60 days.
Anthropic aims to double its computing capacity from 3.5GW to about 7GW next year, enough electricity for every home in Chicago, the NYT reported.
Apple warned investors in its annual filing that AI agents may act in unintended or unpredictable ways, including completing transactions without the right permissions, reports said.
An adviser to the EU's top court said chatbots can infringe publisher rights by reusing protected news, in a case over Google Gemini (the opinion isn't binding).
Anthropic told IPO investors that Claude helps its 15-person chip team design custom silicon faster, a claim based on its own estimates, The Information reported.
Goldman Sachs found workers likely displaced by AI in 2025 took about a month longer to land new jobs than other laid-off workers.
Want absolutely EVERYTHING that happened in AI this week? Click here!
🍪 Treats to Try
Apple's Siri AI Developer Preview lets you expose your app's actions, like booking a reservation or placing an order, to Siri (developer betas only for now).
Netflix's AI Discover starts with a mood or occasion on your TV, then reshapes recommendations as you refine or reject picks (a small U.S. test, English only).
CrowdStrike's AI Kill Switch shuts down an unauthorized AI agent in real time and stops it from running again anywhere in your organization (endpoint enforcement is coming soon).
Flockbay pairs the Godot game editor with an agent that writes your scripts, playtests the game, and patches the bugs it finds —free plan, then $20/mo.
Videco turns a product demo or webinar recording into scored short clips, blog drafts, and social posts, and hosts the video too.
Synter Canvas plans, launches, and tweaks ad campaigns across 30+ platforms from one agent, with approval gates before changes go live.
Testifly explores your web app, writes end-to-end tests, runs them on every deploy or on a schedule, and keeps them updated as your UI changes (use a staging account).
Teamily AI is a group-chat messenger where you and several always-on agents share conversations, documents, meetings, and workflows.
UPCOMING EVENT: The Neuron IRL in San Francisco
The Neuron is going IRL on November 18 at 4:30 PM PT! Join us in San Francisco for drinks, bites, special guests, and a live recording of The Neuron podcast. Huge thanks to Slack and Alumni Ventures for making it happen.
It’s free, but space is limited. Save your spot!
📖 Midweek Wisdom
Anthropic secretly gave an AI agent a coin-flip preference between two equally defensible options, then watched 50 expert analysts run a farm assessment with it. Each row shows how often analysts picked option A when the AI leaned B (purple) versus A (green). The gaps are huge: 82 points on how to measure productivity. Across four key decisions, analysts matched the AI’s lean 86% of the time.
Julia Carpenter reported in The Washington Post that Stanford researchers say AI usage quotas reward activity over value, while skills-based incentives encourage real adoption.
Anthropic researchers ran an experiment in which 50 analysts followed their AI agents' randomly assigned preferences on 86% of decisions, raising doubts about independent human oversight.
Matteo Wong argued in The Atlantic that AI infrastructure spending now props up so much U.S. growth that a crash could threaten the broader economy.
Ethan Mollick asked in a Vox interview which skills we should refuse to automate, since AI can make work easier while quietly weakening abilities we still need.
Carl Benedikt Frey explained in Project Syndicate why AI wins stronger public support in China, India, and Indonesia than in the U.S., tying it to each society's existing economic conditions.
New from The Neuron: AI Explained
Click the image to watch the video!
New episodes air every week on Wednesdays: Spotify | Apple Podcasts | YouTube
A Cat’s Commentary
 | That’s all for now. If you want to get featured above, fill out the poll below and tell us how we did today! |
Btw: We just launched a robotics newsletter! Sign up for it here.
P.S. Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.