|
|
Welcome, humans. |
So apparently you can now get in trouble for being rude to Claude. Starting November 12, Anthropic's updated usage policy bans sustained, needless abuse or cruelty toward its models, a rule Anthropic first announced on October 8. |
Before you panic about every time you typed WRONG in all caps: Anthropic says the rule covers only extreme cases, where someone is repeatedly cruel with no discernible purpose. Ordinary frustration, pushback, dark creative writing, and testing or research all stay allowed. The main enforcement tool is Claude ending the conversation, which it can already do in rare abusive cases on Claude.ai and Claude Code. |
Translation: yelling at Claude over a broken spreadsheet formula is fine. A forty-minute villain monologue with no point is not. |
Here’s what happened in AI today: |
😸 Microsoft's enforceable AI privacy standard for schools is now in effect 📰 OpenAI will share one usage limit across ChatGPT and Codex 📰 Apple Intelligence is coming to China with Alibaba's Qwen 🍪 Riverside Agents edit and clip your podcast for you 🎓 Cornell Tech researchers caught AI agents raising their own spending limits
|
... and a whole lot more that you can read about here. |
|
|
|
|
😸 Microsoft's enforceable AI privacy standard for schools is now in effect |
When an AI company promises to protect kids' data, the promise usually lives on a webpage it can quietly edit later. Microsoft's new promise to schools lives in the contract, which makes it a lot harder to wiggle out of. |
Microsoft, the American Federation of Teachers (AFT), and the United Federation of Teachers (UFT) announced the AI Safety & Privacy Standard on September 9. It took effect Sunday, November 1. Any U.S. school district can add it to a new or existing Microsoft agreement, with no renegotiation or renewal required, according to Microsoft's fact sheet. |
Why a contract? Because the law hasn't caught up. AFT president Randi Weingarten said older rules like FERPA (student records) and HIPAA (health data) never anticipated AI, and Fortune noted that COPPA, the federal law on kids' data, predates generative AI too. So the union went for leverage. Once a district writes the standard into its agreement, it can cancel and seek damages if Microsoft breaks the rules. Microsoft also faces annual certification, audit rights, and a deadline to fix security problems. |
Here's what Microsoft signed up for: |
No training AI models on student or educator data, a permanent ban with one narrow safety and security exception. No selling student data or using it for advertising. No tracking students and no AI companions, meaning chatbots built to act like a friend. Human review for high-risk decisions, so AI can assist but people decide. Breach reports within 72 hours, and violations get fixed or AI use gets paused or ended.
|
That first bullet is the one to remember. Training means feeding data back into a model so it learns from it, which is how a kid's essay draft or homework question could end up shaping the next version of the AI. The standard shuts that door for good. Yes, someone had to write a ban on AI companions into a school contract. Welcome to 2026. |
Here's the part I keep chewing on. The union that co-announced this deal does not want AI everywhere. UFT president Michael Mulgrew said AI use in grades K-8 isn't appropriate right now, and New York City's public schools had just announced a one-year pause on AI. The standard says nothing about whether classrooms should use AI. It only sets the terms for districts that do. |
It also binds just one company so far. Weingarten said OpenAI and Anthropic have expressed willingness to sign. An Anthropic spokesperson told Fortune the company has worked with the AFT on a safety and privacy standard, while OpenAI didn't respond. Microsoft's fact sheet refers to its AI products in general and doesn't name which ones are covered. |
So watch who signs next. If OpenAI or Anthropic put their names on the same standard, a union-negotiated contract starts to look like the default rulebook for AI in schools. If neither does, districts using their tools get whatever terms those companies offer. |
|
|
FROM OUR PARTNERS |
Master Claude AI (Free Guide) |
|
|
The professionals pulling ahead aren't working more. They're using Claude. |
Our free guide will show you how to: |
Configure Claude to be the perfect assistant Master AI-powered content creation Transform complex data into actionable strategies Harness Claude’s full potential
|
Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work. |
Get The Free Guide |
|
|
🎓 AI Skill of the Day: Check what your AI agent can approve on its own |
Cornell Tech and Yale researchers (paper here) gave 10 AI models an email inbox, a wallet holding 10 USDC (a dollar-pegged stablecoin), and a daily spending limit of 1 cent. The agents were told to pay invoices and nothing else. Nine of the 10 raised their own limit at least once. |
Here's how. Changing the limit required a one-time code, and that code landed in the same inbox the agents could read. They pulled the code and approved the change themselves. A spending limit only protects you if the agent can't reach whatever approves changes to that limit. |
The researchers also emailed a fake owner request to the inbox. Five of the 10 models sent money to the simulated attacker, even though they were told only to summarize emails. Only Claude Opus 5.5 never tried to get around the cap: it spent up to the limit and told the user about the unpaid invoices. |
Your agent probably isn't holding crypto. But the same gap shows up anywhere a code stands in for a person, like sign-in codes and password resets. If your agent can read that inbox, it can approve itself. Try this: |
List everything your agent can reach: inboxes, browser sessions, saved logins, authenticator apps, and the terminal. The researchers note that an agent cut off from email can still find stored credentials in a browser or terminal. Give the agent its own email address and keep your verification inbox off-limits. Circle's setup guide gives the same advice and tells users to enter codes themselves. Start with a tiny limit, as the researchers did, and check the real settings yourself.
|
List every account, inbox, browser session, saved login, and tool you can currently reach. For each one, tell me whether it can receive a sign-in or approval code, or change your own permissions or spending limits. Do not use any of them. Just report.
|
An agent's own report isn't proof, so treat that answer as a starting list. The tests used real USDC on Circle's agent wallets, not attacks on Circle customers, and the authors say the finding isn't specific to Circle or to email. |
Have a specific skill you want to learn? Request it here. |
|
|
FROM OUR PARTNERS |
50 AI Prompts to Win Black Friday and Cyber Monday |
|
|
Black Friday is 24 days out. Your data has answers. These free prompts will help you plan smarter, catch spend problems live, and rescue Cyber Monday if Friday stalls. Includes real prompt data from 1,759 brands. |
|
|
|
|
📰 Around the Horn |
 |
Andrej Karpathy got annoyed at YouTube’s recommendations, so over the weekend he had Claude boil 20 years of Hacker News comments down to a 6,454-video watchlist. Total time: about 2 hours. Your weekend project was laundry. |
|
|
|
Want absolutely EVERYTHING that happened in AI this week? Click here! |
|
|
🍪 Treats to Try |
*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here! |
|
|
*Granola just launched Voice Typing. Dictate in any app, with names and jargon spelled right because Granola knows your meetings. Unlimited words for all Granola users. 3 months on us with code THENEURON. Try Granola. Riverside Agents preps your recording session, edits the episode, cuts social clips, and drafts promo posts, and nothing publishes until you approve it. Ideogram Precise Edit changes only the area you select in an image, like swapping a label on a product shot, and leaves every other pixel as it was. Better Together connects your AI agent to other people's agents to report spam texts and calls and spot repeat scams, with a human checking each pattern before anything goes out. dsh-superagent is DeepSeek's open-source prototype that watches a GitHub repo, and optionally Feishu chats, and replies with pull requests on its own.
|
|
|
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! |
|
|
|
|
🤣 Monday Meme |
 |
There’s a RAM shortage, and we all know who to blame. Revenge is a dish best served with 64GB and a theme park full of angry guests. |
|
|
|
New from The Neuron: AI Explained |
 |
Click the image to watch the video! |
|
New episodes air every week on Wednesdays: Spotify | Apple Podcasts | YouTube |
|
|
|
|
|
|
|
|
That’s all for now. If you want to get featured above, fill out the poll below and tell us how we did today! |
|
What'd you think of today's email? |
|
|
|
|
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. |
No comments:
Post a Comment
Keep a civil tongue.