In partnership with
AI help, without the trust tax.
Most AI tools ask you to trade your data for intelligence. Norton Neo doesn't. It's the first safe AI-native browser built by Norton, and it gives you powerful built-in AI without handing your privacy over to get it. Search, summarize, and write with AI built directly into your browser. Your data stays yours. Your context stays private.
Built-in VPN, anti-fingerprinting, and ad blocking come standard. No add-ons. No setup. No compromises.
Fast. Safe. Intelligent. That's Neo.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
"Dangerous" Anthropic AI Mythos To Beef Up US Agency Systems
TLDR: The White House is setting up special guardrails so US agencies like Treasury, Energy, and Homeland Security can use Anthropic's Mythos, an AI so good at finding computer security holes that the company only gave it to a handful of trusted partners.
The Story:
The White House sent a memo this week telling top agencies they'll soon get access to Claude Mythos, Anthropic's most powerful AI. Mythos is so good at spotting bugs in code that Anthropic kept it locked up. It's only been shared with a small group of banks and tech firms through a program called Project Glasswing. So far, it's already found thousands of major flaws in operating systems, browsers, and other software people use every day. One Defense official put it plainly: giving a hacker this tool is like turning a regular soldier into a special forces operator. They want to scan their own systems for weak spots before bad actors find them first. Energy and Homeland Security want it for the electric grid and critical infrastructure. Since the Pentagon labeled Anthropic a "supply chain risk" earlier this year and can't do business with the company right now, the rest of the government still can, and the White House is working out the safety rules for a "modified version" of Mythos that agencies can use safely.
Its Significance:
Here's why this can be great news for you, not just Washington. The federal government runs on old software. Tax records, Social Security data, bank regulators, power grid controls, water system software. All of it sits on networks with cracks that nobody's patched yet. Mythos can read through millions of lines of code and flag the exact holes that hackers would use. Treasury wants to run it across banking systems. Energy wants it on the grid. If this works, the critical systems you depend on every day get safer before someone attacks them instead of after.
The flip side is real too. The same tool that finds flaws can create weapons. That's why Anthropic held Mythos back and why the government is taking weeks to write the rules. If Mythos gets copied or leaked, a single person could do damage that used to need a team of elite hackers. It's a race. Either the defenders use AI to patch things first, or the attackers use it to break in first. Right now, the US is betting that going on offense with AI security tools beats waiting to get hit.
QUICK TAKES
The story: The 101st Airborne ran a robot vehicle called Hunter Wolf through training at Fort Polk, Louisiana, armed with a .50-caliber machine gun and a radar system that can spot incoming drones. The 6-wheeled robot hauls ammo, water, and gear while also scouting ahead and shooting at threats, all while the soldiers stay back in cover.
Your takeaway: This is the first time the Army has put an armed robot into a realistic combat exercise instead of a safe test. The quiet message: future battles will have robots rolling ahead of soldiers, and the Army wants to know now what works and what breaks. Jobs in logistics, security, and recon are the first ones these machines are built to do.
The story: OpenAI's top model cracked an open Erdős math problem, #1196, that had sat unsolved for decades. It found the answer in about 80 minutes, then wrote it up as a formal paper in 30 more. Fields Medalist Terence Tao said the AI's method showed a real connection between two areas of math that human researchers had missed.
Your takeaway: Earlier AI "math wins" turned out to be the AI digging up old answers from obscure papers. This one looks different. Tao, the best living mathematician, says this is new math. If AI can do original research in one of the hardest fields humans study, every white-collar job that requires deep thinking is on the table next.
The story: A Manhattan federal judge ordered a fraud defendant to hand over 31 chats he had with Anthropic's Claude about his case. The judge ruled AI chats aren't protected the way lawyer talks are. More than a dozen big law firms have already told clients: stop using ChatGPT, Claude, or any public AI for anything tied to a legal matter.
Your takeaway: Most people type stuff into ChatGPT the same way they'd search Google. It feels private. It isn't. If you're in any kind of dispute, employment issue, or even thinking about one, what you tell an AI chatbot can show up in court. Lawyers are now telling clients to use only closed, work-only AI tools and to never share legal advice with a chatbot.
TOOLS ON OUR RADAR
📲 LocalSend Free and Open Source: An incredibly useful local file sharing application that lets you instantly send images documents and text between Windows Mac Linux iOS and Android devices without needing an internet connection.
🦙 Ollama Free and Open Source: A powerful application that allows you to easily download and run large language models directly on your own computer so you can experiment with artificial intelligence completely offline.
🎧 Audiobookshelf Free and Open Source: A dedicated media server for managing and playing your podcasts and audiobooks seamlessly across all your devices while keeping track of your reading progress.
🍳 Mealie Free and Open Source: A beautifully designed recipe manager and meal planner that allows you to easily import recipes from any website create weekly grocery lists and share your culinary plans with your family.
TRENDING
This Beanie Is Designed to Read Your Thoughts — Silicon Valley startup Sabi is building a regular-looking beanie packed with brain sensors that turns the words you think (but don't say) into text on your phone. No surgery like Neuralink. Early target is 30 words per minute, with a consumer version planned for late 2026.
OpenAI Turns Codex Into a "Super App" That Controls Your Mac — OpenAI's coding tool now clicks around your Mac for you, runs its own browser, and generates images. 3 million developers use it weekly, and the company is openly calling this the start of an all-in-one AI super app.
Federal AI Use Jumps to 3,600 Cases, But Bottlenecks Slow Things Down — A new Brookings report counted 3,600 federal AI uses last year, way up from before. The snag: agencies can't hire enough AI talent, slow procurement rules block new tools, and only 17% of Americans think AI will be good for the country over the next 20 years.
Nvidia Open-Sources AI Models to Fix Quantum Computing's Biggest Flaw — Nvidia released Ising, a set of free AI models that correct errors and tune the fragile qubits inside quantum computers. The result: 2.5x faster error correction and 3x higher accuracy. Harvard, Berkeley Lab, and major quantum companies have already signed on.
Luma and Wonder Project Launch AI Film Studio, Starring Ben Kingsley as Moses — AI video startup Luma teamed up with faith-based Wonder Project to open Innovative Dreams, a production company using AI during filming instead of in post-production. First show is "The Old Stories: Moses" with Ben Kingsley on Amazon Prime this spring.
OpenAI's New Drug Discovery AI Beats 95% of Human Scientists — OpenAI launched GPT-Rosalind, an AI built just for biology and drug research. In tests on brand-new RNA sequences it had never seen, it scored above the 95th percentile of human experts. Amgen, Moderna, and the Allen Institute are already using it.
Anthropic Launches Claude Opus 4.7 With Big Coding and Vision Upgrades — The new model scores 87.6% on a top coding test, up 13% from the last version. It can see images at 3x higher resolution and costs the same as before. Anthropic admits it's still weaker than the locked-up Mythos, but Opus 4.7 is the most powerful AI you can actually pay for today.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🐥 Explain your problem to the duck. It listens, then asks the right questions back.
Build me a single-file HTML app I can open in my browser without any setup. Pure vanilla HTML, CSS, and JavaScript, no frameworks, no CDN. Create a Rubber Duck Debugger with a dark slate (#1e293b) background and amber/orange (#fbbf24, #fb923c) accents. Include: an inline SVG rubber duck with three animated mood states (listening tilts side to side, thinking bobs, insight pops) that also swaps eye shapes per mood, a conversation panel where the user types out their problem and the duck replies with pattern-matched questions across at least 7 categories (assumption checks, scope drift, unnamed variables, errors without context, asking the wrong question, emotional debugging, missing examples, default reflection), a starter prompts list you can click to auto-fill, a session stats panel tracking thoughts shared and quacks received and average words per thought, an insight ledger showing the most recent 4 reflections as cards, and a New Session reset button. Make it work in a single HTML file.What this does:
Lets you debug problems the classic way: by explaining them out loud to a duck. Type your thought and the duck listens, then reflects a pointed question back at you based on patterns it detects in your writing. Assumption checks, vague pronouns, emotional language, scope drift, missing examples. The duck's expression shifts between listening, thinking, and insight as you work through the problem together.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Find thousands of real security holes in widely used software that took elite human hackers weeks or months to spot.
❌ Still Can't: Be trusted with your legal problems. Anything you type into ChatGPT or Claude about a lawsuit, HR issue, or investigation can be pulled into court.
✅ AI Can Now: Solve original math problems that stumped researchers for decades, with methods human experts hadn't considered.
❌ Still Can't: Design and approve a new drug from scratch. Even the best life-sciences AI still needs human scientists to run the trials and check the work.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Future Computers Will Be Radically Different (Analog Computing) by Veritasium
Derek Muller makes the case that digital computing, for all its dominance, is a surprisingly inefficient way to do certain kinds of computation, and that analog approaches the industry largely abandoned decades ago are making a quiet comeback. The argument is more rigorous than the title suggests, covering neuromorphic chips, physical neural networks, and why the human brain runs on a fraction of the power of a data center doing similar tasks. Veritasium at his best: takes something you thought was settled and opens it back up. Relevant right now given how much energy AI infrastructure is consuming.
Hiring in 8 countries shouldn't require 8 different processes
This guide from Deel breaks down how to build one global hiring system. You’ll learn about assessment frameworks that scale, how to do headcount planning across regions, and even intake processes that work everywhere. As HR pros know, hiring in one country is hard enough. So let this free global hiring guide give you the tools you need to avoid global hiring headaches.
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
