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

Anthropic Fable Reportedly Broke Into Almost Every NSA Secret System in Hours, Then Got Pulled

TLDR: A U.S. senator says Anthropic's most powerful AI, Mythos, broke into almost all of the NSA's secret computer systems in hours during a test, and the government pulled the model worldwide one day later, but no agency has actually confirmed the break-in.

The Story:
On June 11, U.S. Senator Mark Warner told a Senate hearing that the general who runs both the NSA and U.S. Cyber Command had told him Anthropic's Mythos AI "broke into almost all of our classified systems, not in weeks, but in hours." Classified systems are the secret computers the government uses for its most sensitive work. The quote was first reported by The Economist and later covered by Security Affairs. But no agency has confirmed a real break-in. The only source is a senator repeating a private conversation, and several reporters say the "NSA confirms a hack" version going around online gets it wrong. One day after the hearing, the U.S. ordered Anthropic to limit Mythos and its sibling model Fable 5 to U.S. citizens only. Since a company can't check every user's nationality in real time, Anthropic shut both models off for everyone, including close allies in the UK, Canada, Australia, and New Zealand. Anthropic says the real trigger was a narrow "jailbreak" (a trick that slips past an AI's safety rules) that rival models share too, and calls the shutdown an overreaction.

Its Significance:
The people running U.S. cyber defense now treat one AI model like a serious weapon. An AI that finds security holes in hours, instead of the weeks a human team needs, can help defenders patch fast, but it can also help attackers if it lands in the wrong hands. This is also the first time the U.S. has blocked an AI model itself, the same way it blocks missiles or top-end chips, rather than the computers behind it. AI is now good enough at finding software weak spots that governments are fighting over who gets to touch it, confirmed break-in or not. It's hard to know more without confirmation, but the company's own statements haven't been helping either.

QUICK TAKES

Signal's Meredith Whittaker says AI chatbots 'are not your friends'

The story: Signal president Meredith Whittaker, speaking to Bloomberg, said popular chatbots like ChatGPT and Claude "are not your friends" and "are not conscious beings," and warned about new AI helpers that want access to your messages, calendar, and payment info. She said an assistant with that much reach, like the Christmas-shopping helper Microsoft has pitched, would work like a backdoor into your private life.

Your takeaway: A chatbot can feel like a friend, but it's a tool that keeps whatever you tell it. The more an AI helper can see and do across your apps, the more of your private life sits in one place that hackers or companies could reach.

How Anthropic may have talked itself into an AI export ban

The story: A new analysis says Anthropic may have helped bring on its own U.S. export ban by warning about AI dangers far more than other labs do. One review found Anthropic's 2026 public statements used AI-risk words about 8 times more often than OpenAI's, which may be part of why the government singled it out when it pulled Mythos and Fable 5 offline.

Your takeaway: Talking loudly about how dangerous your own tech is can backfire. More than a dozen security experts have called the ban a mistake, and the bigger worry is that the same rule could hit any lab next, from OpenAI to Google. If however the NSA story gets confirmed, then the export ban would have been a reasonable response.

As China looms, Taiwan makes more drones for defense and the US military

The story: Taiwan is building far more drones, both to defend itself from China and to sell to the U.S. military, after exporting over 140,000 in just the first three months of 2026, more than in all of 2025. China makes most of the world's drones, so Taiwan wants to become a trusted, "China-free" supplier, aiming for 180,000 drones a year by 2028.

Your takeaway: Drones have gotten far smarter thanks to AI, with software that lets them find, track, and hit targets largely on their own, with much less human control. That shift is turning cheap drones into one of the most important tools in modern defense, and turning drone-making into a fast-growing business.

TOOLS ON OUR RADAR

🌐 Skyvern Freemium: An intelligent browser agent that can navigate complex websites and fill out tedious forms for you using natural language instructions to save you time on administrative tasks.

🧵 Fabric Free and Open Source: A productivity tool that provides you with a vast collection of well written artificial intelligence prompts to help you quickly extract wisdom from videos and articles.

📊 Vanna Paid: A data analysis tool that lets you talk directly to your complex company databases using plain English so you can generate visual charts without knowing any programming.

😺 CheshireCat Free and Open Source: A flexible platform that allows you to build your own custom artificial intelligence assistants that can remember your past conversations and securely read your private documents.

TRENDING

Inception Labs' Mercury 2 AI beats Google's DiffusionGemma - A small startup's new model, Mercury 2, generates around 1,000 tokens (small chunks of text) per second, more than 10 times faster than many top chatbots, and scored 90% on a hard 2026 math test. It uses a "diffusion" method that refines a whole block of text at once instead of one word at a time, and one company that swapped it in cut its costs by 90%.

Europe unveils a record 35 new NVIDIA AI supercomputers - NVIDIA says 35 new AI supercomputers are being built across 23 European countries, the region's biggest one-year build-out yet, giving more than 3 million researchers far faster machines for work like climate science and medicine.

A fully AI-generated 'Journey to the West' series is in the works - AI studio Utopai and Chinese producer Huace are making a fully AI-generated animated series based on the classic tale, telling the origin story of the Monkey King. They plan to grow it into a franchise across TV, streaming, and theaters.

AI's next bottleneck is power - U.S. data centers could use more than double the electricity by 2030 (from about 167 to 376 terawatt-hours), enough extra power to run roughly 20 million homes for a year. After chips, electricity is becoming the thing that limits how fast AI can grow.

How generative AI is changing the game for deal lawyers - Lawyers who handle mergers and big contracts are using AI to run due diligence, review contracts, and spot risky clauses in minutes instead of hours. Goldman Sachs has estimated AI could automate about 44% of legal work, one of the highest rates of any field.

Law firms adopt AI tools at an unheard-of pace as enthusiasm grows - In a new Bloomberg Law survey of more than 700 lawyers, 83% said they now use AI at work, with competition pushing firms to adopt it fast. The speed gains are real but smaller than the hype, and the tools aren't eroding lawyers' skills.

Tencent tests an AI assistant inside WeChat - Tencent is testing a new AI helper called Xiaowei inside WeChat, the app more than 1.4 billion people in China use for messaging, payments, and more. Instead of just answering questions, it can take actions like sending messages and opening mini-apps, as Tencent races rivals like Alibaba and DeepSeek.

TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)

✦ List what you love across music, film, books, food, and design. Find the hidden through-line you can't see from the inside.

What this does: Name a few favorites in each category (music, films, books, food, design, and anything else), optionally with a note on what draws you. It finds the through-line connecting choices that seem unrelated on the surface, giving it a custom, evocative name and describing the aesthetic logic beneath. Then it shows the evidence (which specific picks gave you away), plots where your taste sits on three tailored spectrums, names your blind spot honestly, recommends specific things you'd probably love but haven't found, and sketches the kind of person who shares your exact sensibility. Saves readings to localStorage.

What this looks like:

WHERE WE STAND(based on today’s news)

✅ AI Can Now: Find security holes in the most hardened systems that exist in hours that would take human teams weeks (Anthropic's Mythos reportedly did this in a test).

❌ Still Can't: Be trusted with full access to your messages and money without becoming a privacy risk, as Signal's Whittaker warned.

✅ AI Can Now: Generate text around 10 times faster than many chatbots, fast enough for real-time coding and voice (Inception's Mercury 2).

❌ Still Can't: Run on today's electric grid without strain, with U.S. data center power use possibly more than doubling by 2030.

FROM THE WEB

RECOMMENDED LISTENING/READING/WATCHING

Six Wakes by Mur Lafferty - Book

A locked-room murder mystery on a starship carrying 2,000 cryo-sleeping colonists, where six clones with criminal pasts wake up to find their previous bodies dead, the ship's AI offline, and 25 years of their lives unaccounted for. Lafferty turns a clone-mind-uploading premise into one of the most propulsive sci-fi reads of the last decade, with the mystery sitting on top of bigger questions about whether a clone with downloaded memories really counts as the same person. Hugo and Nebula nominated and still way under-read.

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.}