Google switches off AI for schools, and Search with it, plus Anthropic's warning

Tuesday, November 3, was election day in the United States, but the AI news kept coming, and most of it was about one question: who gets to switch AI on and off?

Google said its AI tools will be off by default for school students, and that Google Search is part of the same switch, which set off a political backlash within hours. Anthropic, the company behind the Claude chatbot, told investors in its paperwork to go public that it may soon reach its own threshold for AI that speeds up the building of AI. Apple is putting an 18+ rating on dating apps and on AI companions with romantic or sexual content. CrowdStrike, the cybersecurity company, launched what it calls an AI Kill Switch and is buying two AI security startups. And Netflix began testing a tab that recommends shows based on your mood.

Here's what happened, in plain English.

Google switches off AI for schools, and Search

Google makes a lot of the technology that schools run on, from Chromebook laptops to Google Classroom. On Tuesday, in a blog post by Chris Phillips, who runs its education business, Google said its AI products will be "entirely off for K-12 student accounts until school administrators explicitly turn them on." K-12 means kindergarten through the end of high school.

The list covers the Gemini chatbot, Gemini Notebook, Gemini in Classroom and Google Search. Phillips wrote that the goal is to give school leaders control, and the confidence to use AI in classrooms as they see fit. Google says it will share more details with schools in the coming months, before the change rolls out, and TechCrunch reports that it came after consultations with schools, teachers, researchers and policymakers.

The part that drew attention is Search. Phillips told The New York Times that "Search and AI are deeply integrated." Search now includes AI Overviews, the AI-written summaries that appear at the top of the results. So, according to that interview, a school that keeps AI switched off won't have Search either, and students will see a message from their school saying it isn't available.

That leaves schools with a hard choice. As TechCrunch points out, a teacher might not want students leaning on AI summaries for research or homework, but still want them to look up sources. Under this setup, a school that wants Search gets its AI features too.

The reaction was quick. CNBC reports that New York Mayor Zohran Mamdani wrote on X that Google is "trying to use its monopoly power to force AI into our public schools." Senator Bernie Sanders called the policy "extortion." Former Federal Trade Commission chair Lina Khan also criticized it. Google replied on X that "schools are in control," and said AI Mode and AI Overviews in Search will offer step-by-step guided learning through problems for students, rather than quick answers. According to CNBC, that reply didn't directly address whether Search requires AI. CNBC also notes that Senators Richard Blumenthal and Josh Hawley opened an inquiry into Google's push of AI into classrooms last month.

Alongside the policy, Google made Guided Learning available worldwide in Google Classroom. It works like an AI tutor: when a teacher assigns an activity, it acts as a one-on-one study tool built around class material, and teachers see where the class may need help. Google says a pilot of the step-by-step approach in Search starts later this month in select US schools.

Anthropic's warning about self-building AI

Anthropic is preparing to go public, and on Monday it filed an updated S-1, the document a company files with US regulators that lays out its finances and its risks. Most coverage focused on the money. TechCrunch found a warning in the risk factors.

It's about recursive self-improvement, or RSI. That's the idea of AI becoming good enough to speed up the work of building the next, better AI, which could then speed things up even more. Anthropic has a set of internal safety rules called its Responsible Scaling Policy. One of its thresholds is dramatically accelerated progress that can be put down to AI.

In the filing, Anthropic says its internal estimates and expectations for future models, not its current ones, suggest it may meet that definition imminently. It also says: "We have not designated any of our current models as exceeding this threshold." And it admits a blind spot. Its growing use of AI in research might outpace its safety work, and it says it has in the past not spotted, and may in the future not spot, a significant speed-up in its rate of progress until after it has already happened.

Anthropic writes that using AI to build more capable AI could also pose a considerable safety risk. It says it doesn't expect to be able to safely use Claude to fully design and build later generations of its most advanced models without meaningful human supervision, and that it is considering updates to its Responsible Scaling Policy to keep humans in control.

TechCrunch adds a caution: an S-1 is both a disclosure and a sales document, so a claim like this can double as marketing. Either way, a company preparing to sell shares has now told investors, in writing, that this moment may be close.

Apple's 18+ rule for AI companions

Apple told developers on Monday that it is tightening App Store rules for younger users, TechCrunch reports. Apps with dating or romantic relationship features need an 18+ rating, and that includes AI companion apps with sexual or romantic content.

An AI companion is a chatbot built to act like a friend or a partner. TechCrunch names Character.AI and Replika as examples, and says apps like these are often aimed at younger people. It also notes that top AI companion apps had been downloaded 220 million times worldwide as of summer 2025.

Starting November 16, developers have to add new content descriptors, which are labels that say what's in an app, so that its age rating matches what it actually does. Apps also have to disclose whether they share a user's real-world location with another person or group, which TechCrunch notes raises obvious privacy concerns when people don't realize it's happening. There's one local exception: in Brazil, family location-sharing apps can seek an official rating from the Ministry of Justice and Public Security.

The change follows Apple's global rollout of age checks and more parental controls. For parents, an app's age rating should now say a bit more about what the app can do.

CrowdStrike's off button for AI agents

More companies are putting AI agents to work. An AI agent is AI that doesn't just answer questions, but takes actions on its own, like sending emails or changing files. That raises an obvious question: what happens when one goes wrong?

CrowdStrike, one of the largest cybersecurity companies, announced the AI Kill Switch at its Fal.Con Europe conference, CNBC reports. It lets a company's security team shut down a misbehaving agent, stop it from restarting and cut off its network access. It's part of Falcon Guardian, CrowdStrike's product for watching the AI agents running inside an organization, and the company says it's available now.

CEO George Kurtz told CNBC's Jim Cramer: "Everyone is building AI agents, but they never had the off button, and now they do." He compared it to airbags in a car, a safeguard you want even when everything is supposed to work, and said that if agents step outside their guardrails while running, CrowdStrike can shut them down.

CrowdStrike is also spending on the same idea. TechCrunch reports that it's buying two startups. Fiddler AI makes software that monitors what AI agents do inside companies, and will become part of CrowdStrike's platform. Dreadnode automates security testing to find flaws in AI models, apps and agents before they go live, and will feed its findings into Falcon Guardian. Citing The Wall Street Journal, TechCrunch puts the prices at about $400 million for Fiddler AI and about $95 million for Dreadnode, close to $500 million in total.

The bet behind both moves: as companies hand more work to AI, they'll pay to watch it, test it and stop it when needed.

Netflix asks what mood you're in

If you've ever scrolled Netflix for half an hour and then watched nothing, this one is for you. Netflix is starting to roll out a new tab called AI Discover, TechCrunch reports. Netflix describes it as a test kitchen for new ways to find something to watch.

The first experiment works by mood. You pick a theme, such as "something light," "scare me," "family movie night" or "quick watches," and you get matching picks with clips. You can narrow things down, or choose "something different" to start over. A new "Tune Your Recommendations" button next to Play and My List lets you save titles you like or rule things out.

Netflix says the feature runs on what it calls a new intelligence layer that responds to what you tell it in real time. For now it's rolling out in the US to people who use Netflix in English on supported TVs, with more people getting it in the coming weeks. Netflix says feedback will decide which ideas make it into the main app.

The big picture

Tuesday's stories share a theme: control. Google is deciding how schools turn AI on and off, and its choice to bundle Search with it is what made people angry. Anthropic is telling investors that it may soon need stronger rules to keep humans in charge of how its own AI gets built. Apple is deciding who is old enough for an AI companion. CrowdStrike is selling an off button for AI agents. Even Netflix's experiment is about letting viewers steer the algorithm instead of just scrolling.

None of this settles the bigger questions. Google says more details will come before its change rolls out, Anthropic says no current model has crossed its line, and Netflix's tab is still a test. But the question of who holds the switch, and how easy it is to reach, is showing up everywhere AI does.