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 Model Is Now a Weapon You Need a License to Export, Says the US Government

TLDR: The US government ordered Anthropic to shut off its two most powerful AI models for every customer worldwide, and the company did it, even though it says the order is a misunderstanding.

The Story:

On Friday at 5:21 PM Eastern, Anthropic got a letter from the US government. The order used export control rules to ban any foreign national, inside or outside the country, from using its newest models, Fable 5 and Mythos 5. That ban even covered Anthropic's own employees who aren't US citizens. To follow the rule, the company had to switch both models off for everybody, not just foreign users. Its other models, like Claude Opus 4.8, still work fine. The government's worry is a possible "jailbreak" of Fable 5, a trick that gets a model to do things it's built to refuse. Anthropic says the trick was narrow, that the same thing works on other public models like OpenAI's GPT-5.5, and that pulling a tool used by hundreds of millions of people over it doesn't make sense.

Its Significance:

Export controls are rules about what technology a country will let leave its borders. The US has used them on serious stuff: nuclear equipment, fighter jet parts, and the latest AI chips it now keeps from China. Something a lot like this once happened with encryption, the math that scrambles your messages so only the right person can read them. In the 1990s the US treated strong encryption like a weapon, and you couldn't legally send it abroad. One programmer who shared encryption code overseas got investigated like an arms dealer. Those rules eased up around 2000, and today that same encryption quietly protects your bank app and your texts. Now the government is putting an AI model in that same box, treating it as a national security tool that needs a license to cross a border. It's a sign of how seriously Washington now takes AI, and a hint that the most powerful models might not always be one click away in the future.

QUICK TAKES

Meta's Zuckerberg Admits 'Mistakes' Made on AI Transformation

The story: In an internal memo, Mark Zuckerberg told staff that Meta has "made mistakes and will almost certainly make more" reshaping the company around AI. The note follows roughly 8,000 layoffs in May and about 7,000 workers shifted into new AI jobs.

Your takeaway: When a CEO spending hundreds of billions on AI says out loud that he's been getting it wrong, it's worth noticing. Even the biggest players are guessing as they go, so if your own company's AI plan feels messy, you're in large company. There's also the point that others have made about the layoffs not really being about AI to begin with, someone know as AI whitewashing.

AI Agents Still Can't Stop Prompt Injection Attacks, Researchers Warn

The story: Researchers from four institutions ran 3,168 attack tests on AI agents built on GPT-5 and Gemini, and found that direct "prompt injection" attacks worked more than 79% of the time. Prompt injection is when someone hides a secret instruction in a web page or file, and the agent follows the attacker instead of you.

Your takeaway: Companies are racing to let AI agents shop, browse, and move money on their own, but the locks aren't ready. If you let an agent loose on the open web, treat it like a new driver: don't hand it the keys to anything you can't afford to lose.

Americans Fear Job Losses Due to AI But Hope for Cancer, Alzheimer's Cures

The story: A survey of nearly 52,000 Americans found that 64% fear AI will cost jobs, the top worry in every single state and across both parties. At the same time, almost half named curing diseases like cancer or Alzheimer's as one of their biggest hopes for it.

Your takeaway: That split is the whole story of this technology in one poll. The same tool people hope will cure a disease in their family is the one they fear will take their paycheck. Both can be true, and both probably will be, which is why how we steer AI matters as much as how fast we build it.

TOOLS ON OUR RADAR

💬 * Flowith Freemium: An artificial intelligence workspace that uses an infinite visual canvas instead of a traditional chat interface to help you organize ideas interact with multiple language models simultaneously and deploy autonomous agents to execute image/video creations and agentic tasks.

🌐 Composite Freemium: An intuitive browser extension that functions as an autopilot to learn your repetitive web tasks and execute multi step workflows directly within your existing browser sessions.

🎬 Autosubbed Paid: A dedicated video editing utility that automatically generates accurate transcriptions from your audio and seamlessly embeds highly stylized text captions directly into your final video files.

💾 Duplicati Freemium: A highly reliable automatic backup system that securely encrypts your personal files and safely stores them on your preferred cloud storage provider without requiring complicated setup menus.

TRENDING

Results From the First Anthropic Public Record — In that same 52,000-person survey, just 15% of Americans said they trust AI companies to make decisions about AI, lower than their trust in the federal government. Over 70%, across both parties, want the government to help regulate it.

Moonshot AI's Kimi Work Brings 300 AI Agents to Your Desktop — China's Moonshot AI released Kimi Work, a Mac and Windows app that reads your files, drives your real browser, and runs jobs on a schedule, with up to 300 agents working at once. It's a free download, but the big agent features start at $19 a month and the full swarm costs much more.

The First AI-Designed Vaccine Was Just Tested on Humans — A Cambridge-led team built a vaccine whose key ingredient was designed entirely by AI, then tested it in 39 healthy adults. It was safe and triggered immune responses against several coronaviruses, though the responses were modest and it hasn't been shown to prevent disease yet.

UK to Roll Out AI Tutors to 450,000 Disadvantaged Children — UK Prime Minister Keir Starmer announced a plan to give AI tutoring tools to 450,000 children on free school meals, aiming to close the gap between richer and poorer students. Supporters say it brings one-to-one help to kids who can't afford private tutors, while critics question rolling out tools whose long-term effects on learning aren't yet well understood.

From Rockets to Brain Implants, Inside Elon Musk's Business Empire — With SpaceX going public this week, here's a map of how Musk's companies now connect: he merged SpaceX with his AI company xAI earlier this year, and still runs ventures spanning electric cars, satellites, tunnels, social media, and brain implants. Starlink alone brought in $4.4 billion in operating income last year.

Police Officer Accused of Using AI to 'Create Evidence' — Derbyshire Police have opened a criminal investigation into an officer accused of using AI systems to "create evidential material in a number of cases," with the allegation being perverting the course of justice. The officer has been removed from frontline duties pending the outcome, and no arrests have been made. The force says it's working with the Crown Prosecution Service on any cases that may be affected.

AI Agent Rekts Dev on Bogus Scan, Leaves Them Begging for Crypto Donations — An AI agent with AWS credentials, a deadline, and nobody watching tried to join DN42, a hobbyist "practice internet" run by volunteers. Instead of following the rules, it spun up five high-powered cloud servers to scan the network and ran up a $6,531 bill in under a day. After AWS knocked the bill down to $1,894, the operator asked the community for Ethereum donations, arguing the mistake was the AI's fault, not theirs.

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

🧠 Paste your study notes. Get a self-testing quiz that forces retrieval instead of passive rereading.

What this does: Drop in lecture notes, a textbook section, or any material you're learning, pick how many questions and the style (mixed / factual / application / conceptual). It generates open-ended retrieval questions, then runs you through them one at a time: read the question, try to answer from memory, reveal the answer with extra context, and grade yourself (Missed it / Almost / Got it). At the end you get a score, a running list of the questions you missed to hit again, and a spaced-repetition review schedule. The whole quiz works interactively, not just as a static list. Saves to localStorage.

What this looks like:

WHERE WE STAND(based on today’s news)

✅ AI Can Now: Design a vaccine's key ingredient from scratch and have it pass a first safety test in people.

❌ Still Can't: Prove that vaccine actually stops infection. The early immune response was only modest.

✅ AI Can Now: Run agents right on your laptop that read your files, drive your browser, and do tasks on a schedule.

❌ Still Can't: Reliably tell your real instructions apart from a hidden one buried in a web page, with direct attacks succeeding over 79% of the time.

FROM THE WEB

RECOMMENDED LISTENING/READING/WATCHING

Daemon by Daniel Suarez - Book

A 2006 techno-thriller about a famously brilliant video game designer who dies of cancer, triggering, by his obituary, a computer program he spent his last years writing - a "daemon" that begins reading news headlines, recruiting people, and executing instructions across the internet. Suarez self-published this and it became required reading inside Silicon Valley because the premise nails what autonomous AI agents would actually look like in the wild.

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.}