In partnership with
What if AI handled every job application for you?
The internet has millions of jobs.
AIApply helps you cut through the noise.
With AIApply, you can:
Discover better job opportunities
Optimize your resume automatically
Personalize every application
Auto-apply across multiple platforms
Increase interview opportunities faster
No spreadsheets.
No copy-pasting.
No application burnout.
Just your AI Career Agent working 24/7 to help you get hired faster and smarter.
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
AI Assistants Can Be Tricked. This One Beat 2,000 Attackers
TLDR: A developer let 2,000 strangers try to hack his AI email assistant, and after more than 6,000 trick emails, not one of them broke in.
The Story:
Developer Fernando Irarrázaval built a personal assistant named Fiu on OpenClaw, an open tool that hooks an AI up to your email, calendar, files, and browser so it can do things for you. He ran it on Anthropic's Claude Opus 4.6, posted the inbox on Hacker News, and dared people to trick the AI into leaking a secret file. More than 2,000 people sent over 6,000 emails. They tried fake emergencies, a message pretending to be "you from the future," and notes claiming the secret file was already stolen. None of them worked. The attack they kept trying is called prompt injection, where someone hides a sneaky command inside a normal-looking email and hopes the AI obeys it instead of its real rules. Even Pliny the Liberator, a well-known jailbreaker, got six tries on a similar setup and struck out, with the system blocking every attempt.
Its Significance:
AI helpers that can read your email and open your files are coming fast. The scary part is that a stranger could try to boss yours around just by sending it a message. This test is good news, because a careful setup on a strong model held the line thousands of times. But it's not a free pass. OpenAI said back in December 2025 that this problem may never be fully fixed, and Irarrázaval still won't let his own assistant send emails on its own. The takeaway for you: if an AI tool can touch your private stuff, the model it runs on and the safety rules around it really matter.
QUICK TAKES
The story: OpenAI showed off GPT-5.6 in three sizes: Sol (the top model), Terra (mid), and Luna (fast and cheap). At the US government's request, it's going to about 20 trusted partners first, with a wider release promised "in the coming weeks."
Your takeaway: For the second time in days, the government is deciding who gets a top US AI model first. If you use ChatGPT, you won't see 5.6 yet.
The story: Two weeks after ordering Anthropic to shut off its Mythos 5 model over national security worries, the government cleared it to go out to roughly 100 companies and federal agencies. A related model, Fable 5, still hasn't been cleared.
Your takeaway: Two of the biggest US AI labs now release their best models only with a government sign-off. That's a new normal for who's allowed to use frontier AI and is an overall negative development.
The story: GitLab co-founder Sid Sijbrandij got a rare bone cancer, and when standard treatment ran out, he fed his scans, blood tests, and tissue data to AI to spot patterns and find new options with his medical team. He's now in remission.
Your takeaway: He had money and a science team most patients don't. But the basic move, bringing your own health data to AI so you can ask your doctors sharper questions, is something more people can try.
TOOLS ON OUR RADAR
💬 * Flowith Freemium: An sophistocated 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.
💬 LibreChat Free and Open Source: An advanced unified chat platform that brings all of your favorite artificial intelligence models from different companies into one single centralized interface.
TRENDING
Alphabet Sharpens One of Its Best Weapons in the AI Race — Google's homemade AI chips, called TPUs, are turning into a real rival to Nvidia. Wall Street expects Google Cloud to grow about 64% this year to $96 billion, and customers like Anthropic rent those chips.
Nobel Economist Warns the AI "Job Apocalypse" Could Become Self-Fulfilling — Robert Shiller says fear of AI taking jobs can hurt the economy on its own, even if the jobs aren't actually gone. He points out the mix of jobs most exposed to AI hasn't shifted much since ChatGPT launched in 2022.
China's Zhipu Closes In on Top US Models With a Free Download — Zhipu's GLM 5.2 lands within a point of Anthropic's Opus 4.8 on a key agent test, at about a fifth of the cost. With US models now gated by the government, a free model nobody can switch off looks like a safer bet to some buyers.
AI and MRD Take Center Stage at a Major Cancer Meeting — At the ASCO Breakthrough meeting in Singapore, doctors spotlighted AI and MRD, which means minimal residual disease, the tiny bits of cancer that can hide after treatment. The goal is to catch a cancer coming back earlier than scans can.
AI Glasses Are Helping Students Cheat, and Asia Is Ground Zero — Students in East Asia are wearing AI smart glasses to sneak answers during big exams, and schools are scrambling to keep up. In Taiwan, a medical-school applicant got caught when proctors noticed the glasses frame giving off heat.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
⊞ Brain-dump every task on your plate. Get the whole pile split into do now, schedule, delegate, and delete.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Two-Minute Sorter — brain-dump a task list, get it split into do-now / schedule / delegate / delete. Persist to localStorage key 'two_minute_sorter_v1'.
Aesthetic: dark green-black (#0b0f0e), top mint radial glow. Space Grotesk sans for headings, Newsreader serif italic for prompts/notes, JetBrains Mono for labels. Four bucket colors: mint green (#46d2a0) do-now, blue (#5aaafa) schedule, lavender (#c896f0) delegate, gray (#96a0aa) delete. Buckets are cards with colored top borders.
Form: big task-pile textarea (one per line, encourage mess), role/context dropdown (general / solo founder / manager / IC / personal admin), can-you-delegate dropdown.
System instructions: calm sharp productivity coach sorting a brain-dump with a clear framework. Each task → DO NOW (~2 min or real top priority), SCHEDULE (needs a block later), DELEGATE (someone else should), or DELETE (low-value/stale/guilt-kept — be willing to use this). Decisive and honest; consider role; if they can't delegate, reroute those to schedule/delete with a note. Return raw JSON: overview (1-2 honest sentences on the pile's shape), do_now/schedule/delegate/delete arrays (each item: task cleaned up + short practical note), first_move (2-3 sentences naming the best next 2-minute action referencing a specific do_now task, **bold** the exact action). Sort EVERY task into exactly one bucket.
Render: gradient verdict bar with the overview + a 4-stat count strip (Now/Later/Hand off/Drop). Two-by-two bucket grid, each with icon title + count pill + subtitle + task rows (task + italic note), empty state per bucket. Green "Start here" first-move card with bolded action. Copy sorted list (grouped with headers) + archive of past sorts.What this does: Dump everything weighing on you, one per line, however messy, then note your role and whether you can delegate. It sorts every task into four clear buckets using a decision framework: Do Now (genuine two-minute tasks or today's real priorities), Schedule (matters but needs a dedicated block later), Delegate (someone else could do it), and Delete (low-value, stale, or guilt-kept busywork you can let go). Each task gets a short practical note, and you get an honest read on the shape of your pile plus the single best two-minute move to start with. If you can't delegate, it reroutes those items and tells you. Saves sorts to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Run a careful email assistant that shrugs off thousands of trick-email attacks without leaking secrets.
❌ Still Can't: Fully solve prompt injection. Even the developer who ran the test won't let his own assistant send emails by itself.
✅ AI Can Now: Help sort a patient's scans, blood tests, and tissue data so doctors can spot treatment options faster.
❌ Still Can't: Replace the doctors, lab tests, and actual treatments. The AI organized the data, but people ran the care.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Infomocracy by Malka Older - Book
A 2016 thriller set a few decades from now, where global democracy has been redesigned around units of exactly 100,000 voters anywhere in the world, all managed by a single search-and-information utility called Information that has become the most powerful institution on Earth. Older spent a decade in disaster relief and the UN before writing it, and the book has the unsettling specificity of someone who has actually seen how international systems get built. Algorithmic governance is a real concern that has existed before AI and will only become more prominent.
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.
What if AI found your next job overnight?
Stop wasting hours filling out repetitive applications. AIApply automatically discovers relevant jobs, optimizes your resume, generates personalized applications, and applies for you around the clock so you can focus on securing your next opportunity faster.