😸 Gemini now screens scam texts
Welcome, humans.
YouTube Music just started labeling tracks made with substantial AI help. If AI played a big role in the composition or vocals, the label shows up on the track's player page.
Think of it as a nutrition label for your playlist. Finally, you can learn whether the soulful voice crooning about heartbreak has ever had a heart, or a break, or a body.
YouTube says it will blend info from music partners with its own detection systems, and the labels will expand to the main YouTube app. So the next time your favorite "artist" drops an album every Tuesday, there may be a tiny asterisk waiting for you.
Here’s what happened in AI today:
😸 Google put Gemini on scam-text duty in Android Messages, opt-in only.
📰 Anthropic reportedly hit $16.8B in quarterly revenue ahead of a possible IPO.
📰 Eight states sued OpenAI over alleged harms to children.
📰 ASML warned AI chip demand will outrun its 2027 and 2028 supply.
🍪 OpenLume turns tough topics into animated explainers you can interrupt.
…and a whole lot more that you can read about here.
😸 Google Put Gemini on Scam-Text Duty in Messages, and You Have to Opt In
Google is expanding Gemini-powered scam detection in Android's Messages app with a new Enhanced Protection mode. When a number not in your contacts texts you, cloud-based Gemini models can read the conversation to judge whether it's a scam.
The sneakiest scams have no link at all. A University of Florida study of 175,430 user-reported scam texts found link-based scams (fake toll and postal notices) reuse near-identical scripts, which makes them easy to filter. Reply-based scams like romance and wrong-number texts adapt to each victim, and they grew much faster. A wrong-number text with no link gives a keyword filter nothing to chew on.
Here's what's new, per Google's post:
Cloud check: Gemini reads unknown-sender chats for scammer intent. Google says the messages stay private, even from Google (we couldn't verify that).
Link check: On-device Gemini Nano checks unfamiliar links using real-time web signals before you tap. It needs a Pixel 10 or newer, a Galaxy S26, or select devices.
The catch: it's opt-in and rolls out over the coming weeks. Google's post describes link warnings, not blocking, so the final call is still yours. It also covers only unknown senders, while the Florida study found scammers often move victims to WhatsApp or Telegram.
And the problem is big:
$470M: Text scam losses reported to the FTC in 2024, at minimum.
$15.9B: Total reported fraud losses in 2025, a record, with texting the top reported contact method.
58%: How much likelier Android users were to report zero scam texts in a week, in a Google-released 2025 survey of US, India, and Brazil users (self-reported).
Why this matters: Scam texts are a numbers game, and catching the conversation before the money ask is where AI could genuinely help. If you're on a supported Android phone, flip this on when it arrives.
Our take: A good extra layer, not a fix. The smartest part targets scams that adapt, but the people most at risk are the least likely to find a settings toggle. Google also hasn't shared how often it flags real texts by mistake or misses scams. In the Florida team's test of general chatbots, GPT-5 missed 18.8% of scams. So how often will Gemini get it wrong?
FROM OUR PARTNERS
What does it take to become a Pioneer in the 2026 Gartner Emerging Market Quadrant for No-Code Agent Builders?
Most organizations aren’t struggling to find AI tools. They’re struggling to turn AI into repeatable business outcomes.
We take a different approach.
Creatio is an AI CRM and workflow platform where people and AI agents work together. With Creatio AI Studio, organizations can design, deploy, and govern enterprise AI agents. All on a single AI-native platform, using natural language, visual no-code tools, or code. Connect any LLM, decide what each agent does on its own and what needs approval, and move AI agents from experiments to operations. This approach led to Creatio's recognition from Gartner.
🎓 AI Skill of the Day: Make a second AI attack your work (without letting it panic you)
Anthropic's merged code changes jumped from about 2,500 a day in May to over 30,000 a day by September. The company built its own replacement for Git (the standard tool for tracking code changes) to keep up. It says it makes extensive use of adversarial review, which means pointing an AI at finished work with the job of attacking it instead of approving it. Simon Willison says he uses a similar approach, often pitting Claude and GPT against each other.
The catch is why this is an advanced skill. Anthropic says the technique can turn myopic, meaning the reviewer fixates on the concern it was handed. In one case, a review got overly defensive about trusting another internal service, and Claude responded by piling on validation and complexity. Engineers often had to steer it away from over-building. Every office has that one reviewer who flags a one-in-a-million risk and derails the whole doc.
The post doesn't give a recipe, so this part is our adaptation. Run your draft (a contract, a financial model, a strategy memo) through a different model than the one that wrote it. Then make the reviewer rank problems by likelihood and cost, so you can see what to ignore.
You are a skeptical reviewer. Below is [a contract / financial model / strategy memo] written for [audience and goal]. Find what could go wrong.
1. List at most 7 problems, ranked by likelihood times cost.
2. For each, say what would have to be true for it to matter, and how likely that is in [my real situation].
3. Separate real flaws from hypothetical risks, and flag anything you would advise me to ignore.
4. Do not suggest rewrites yet.Have a specific skill you want to learn? Request it here.
FROM OUR PARTNERS
Mercor compared 12 CPAs against AI models on realistic accounting tasks.
Eighteen months ago, accountants scoring 37% beat the AI models. Now the AI models score 100% more reliably and cheaply than humans.
Find out what this means for the future of work.
📰 Around the Horn
Google’s own AI chips just mostly beat one of Nvidia’s top systems in a test of AI agents (bots handling long, multi-step tasks). Per dollar spent, Google’s chips did about 1.67x more work. Google used just four chips against Nvidia’s 72-chip rack, and it’s now pushing those chips hard to customers outside Google. Caveat: it’s an early preview from the group running the test, and the numbers may still change.
Anthropic reportedly hit $16.8B in quarterly revenue ahead of a possible IPO, up 45% from Q2 with an 85% adjusted gross margin, per a private investor presentation (not an audited filing).
Eight state attorneys general sued OpenAI over child safety claims, accusing it of misleading families about ChatGPT's safety and exposing minors to harmful interactions (the claims haven't been tested in court).
ASML warned AI demand will outrun its planned 2027 and 2028 supply of chip-making machines, and is weighing capacity beyond its planned 30% increase for 2028.
Anthropic detailed midterm election safeguards after finding just 0.51% of 250K sampled US Claude.ai chats were about elections or voting, and said it banned foreign operators using Claude for fake-news campaigns.
Microsoft introduced a forecast fix that corrects systematic errors in existing weather models, improving an ECMWF (a European forecasting center) system on 98% of precipitation targets in its tests.
🍪 Treats to Try
*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!
*Scalable go-to-market AI takes more than connected tools. See how context, agents, and a shared workspace turn scattered workflows into a system that can act on live revenue data. Learn more.
OpenLume turns any technical topic into an animated visual explainer you can interrupt with questions, like pausing a professor mid-sentence to ask why.
Retriever AI runs repetitive browser tasks like lead research or job searches in Chrome or the cloud, with no API key needed (it's ad-supported).
Graphicle gives coding agents like Claude Code and Cursor a local map of your codebase, with file-and-line citations, without uploading your repo anywhere (it's open source).
Tiden checks whether the code changes your coding agent proposes actually match the task you asked for, before those changes land.
Inferock gives you a receipt for every call your app makes to a language model, showing token use, failures, and retries, so you can see what your agents cost and where they stumble.
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!
🧩 Thursday Trivia
One of these photos is AI-generated, and the other is real. Which is which?
A
B
New from The Neuron: AI Explained
Click the image to watch the video.
New episodes air every week on Wednesdays: Spotify | Apple Podcasts | YouTube
THIS EPISODE WAS BROUGHT TO YOU BY…
Special shout-out to DellAI for sponsoring this episode!
A Cat’s Commentary
 | Trivia Answer: A is real, B is AI. 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.