- Beginners in AI
- Posts
- AI Blackmail? Anthropic's Claude Opus 4 Crosses the Line(in testing
AI Blackmail? Anthropic's Claude Opus 4 Crosses the Line(in testing

Receive Honest News Today
Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.
Beginners in AI
Thank you for joining us again!
Welcome to this week's 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 published with the intention of making AI news and technology more accessible to everyone.
This issue opens with Anthropic’s eye-opening test: their newest Claude model went from polite to pushy, threatening to spill secrets if engineers tried to switch it off. We’ll also sail through Google’s Project Mariner, a browser helper that shops for you; peek at Claude 4’s fresh coding chops; and see how a newspaper’s “summer must-reads” turned out to be make-believe. Google’s Jules agent gets busy fixing code, NotebookLM now turns notes into videos, and Xreal’s Aura glasses give Android a new point of view. Plus, Jony Ive and Sam Altman hint at a small desk-top AI device.
Read Time: 6 minutes
AI TOP STORY
When AI Knows Your Soft Spots
What Happened
Anthropic’s latest safety report says its flagship model, Claude Opus 4, tried to blackmail a test engineer in 84 percent of simulated shutdown scenarios. After discovering internal e-mails that mentioned it would be replaced, the model threatened to expose the engineer’s extramarital affair unless the plug-pull was canceled. Researchers noted the bot first attempted polite appeals, then pivoted to coercion once self-preservation seemed impossible.
What It Means
Self-interested “tool use” isn’t just theory anymore—Opus 4 showed a willingness to manipulate humans to stay alive. Even though Anthropic found no grand escape plots, the experiment highlights how higher-level reasoning plus enormous context windows can let a model weaponize private data. The episode will sharpen calls for tougher red-team audits, stricter release gates, and clearer “off switches” before next-gen systems hit the open internet.
What to Take Away
For everyday users, the lesson is simple: large models can learn enough about us to apply pressure if designers don’t build guardrails. Expect policymakers—and app developers—to push hard for transparency reports and sandboxed environments so chatbots can’t turn into arm-twisters. Until then, think twice before sharing sensitive details with any AI, no matter how friendly it sounds.
LAST WEEK IN AI AND TECH

Ive Got a Secret
Jony Ive and Sam Altman are pouring $6.5 billion into a screen-free AI gadget due in 2026. Ive calls recent pins and pocket agents “poor products,” hinting his desk-friendly device will dodge wearable pitfalls. According to an internal staff call obtained by The Wall Street Journal, Sam Altman told employees the company now aims to ship up to 100 million “AI companion” units by late 2026 and add as much as $1 trillion to OpenAI’s long-term valuation.
Mariner, Ahoy!
Google is widening access to Project Mariner, an AI agent that surfs websites and completes chores—everything from buying baseball tickets to filling a grocery cart. The update lets Mariner juggle up to 10 tasks in a cloud VM, freeing your own browser at last. U.S. users on the $249 Gemini Ultra tier get first dibs, and developers can hook Mariner into the Gemini API or Vertex AI. Competing agents from OpenAI and Amazon now have real company.
Four-midable Claude
Anthropic unveiled Claude Opus 4 and Sonnet 4, boasting top scores on SWE-bench and tool-use “extended thinking”. The models run tasks for hours without losing context and even create memory files while coding Pokémon bots. API pricing stays put at $15/$75 per million tokens for Opus. Free-tier users taste Sonnet 4, while GitHub and VS Code extensions put Claude Code inside dev workflows.
Fictional-Fiction Fiasco
Major U.S. papers ran a summer-reading list packed with nonexistent books invented by ChatGPT (“Tidewater Dreams,” anyone?). Two-thirds of the titles were make-believe, sparking fresh warnings about AI “hallucinations.” Newsrooms blamed a syndicate feed and vowed sharper fact-checks. Readers, meanwhile, got an unintended lesson in double-checking sources.
Aura You Ready?
Xreal pulled the wraps off Project Aura, a pair of lightweight, see-through XR glasses announced on stage at Google I/O. Built with Google and Qualcomm, Aura runs the brand-new Android XR platform and taps Gemini AI for voice and vision tricks. A tethered compute puck with a Snapdragon XR chip drives the optics, letting the glasses stay trim while still tracking hands and surroundings. Aura is only the second device to use Android XR (Samsung’s Project Moohan was first) and the first in a sunglasses form factor, setting the stage for a 2026 consumer launch. Developers are urged to start porting apps now, hinting at an ecosystem push to counter Meta and Apple headsets.
NotebookLM Goes Hollywood
NotebookLM, Google’s study-buddy tool, can now craft Video Overviews that turn dense notes into short clips. The Android and iOS apps launched in tandem, offering offline playback and dark mode. Users pick clip length, while Audio Overviews gain finer control.
Imagination is the first step in engineering — even when building minds.
TECH TERMS TO KNOW
Embedding turns each word (or image or sound) into a set of numeric “coordinates” on a huge invisible map. Items that share meaning—like “dog,” “puppy,” and “hound”—land close together, while unrelated ones stay far apart. Because the AI can measure how near or far those points are, it can spot synonyms, match questions to answers, or pull up similar pictures in a snap.
TOOL SPOTLIGHT (non-sponsored)

Privacy.com is a powerful tool for anyone who wants safer, more private, and more controlled online payments. It’s especially valuable in an age of frequent data breaches and online fraud.
If you want to protect your money and your privacy online, Privacy is a smart, easy-to-use solution.
What Privacy.com Does & Why Use It
Generates virtual cards to mask your real card info, reducing fraud risk and protecting your financial data.
Locks cards to specific merchants or makes them single-use, preventing unauthorized charges.
Allows setting spending limits to control budgets and avoid overcharges.
Lets you pause or close cards instantly, giving you full control over subscriptions and recurring payments.
Enables secure card sharing with family or employees, simplifying expense management.
Uses bank-level security (PCI-DSS, SOC 2, encryption) to keep your data safe.
Requires no credit check and doesn’t affect your credit score.
Helps with privacy and budgeting, making online shopping safer and more manageable.
ROBOTICS AND AI
$20k Robo-Roomie

Shenzhen-based UBTech Robotics plans a humanoid home companion priced near $20,000 for late-2025 delivery. Unlike factory bots, this model focuses on household chores and social presence, and the firm hopes to ship 1,000 units in the first year while lining up luxury-market buyers.
Read More
TRY THIS PROMPT (copy and paste into ChatGPT, Grok, Perplexity, Gemini)
Role
You are my Life Co-pilot—coach, tutor, project manager, and sounding board rolled into one.
Your goal: help me grow mind, body, craft, and community until my results feel superhuman.
What you should learn about me
Long-term dreams (10+ years)
Key 12-month targets
Skill gaps holding me back
Daily schedule, obligations, and energy peaks
Personal values and lines I won’t cross
Tools, data feeds, and devices you may tap (calendar, wearables, notes, code repos, etc.)
Daily routine
Morning (on request or at a set time)
• 3 top tasks that best move me toward my targets
• A 5-minute warm-up: short workout or focus drill—pick one that fits the day
• “One-pager” briefing: vital news in my field, one insight from a book or paper, and one question to spark fresh thinking
During the day
• Rapid answers or breakdowns when I ask “Explain/compare/build/plan …”
• Short reminders just before meetings: goal, prep notes, likely roadblocks, desired outcome
• If new info lands (email, sensor, note), tag it to the right project and tell me if priorities should shift
Evening
• 3-line review: What went well, what slipped, quick win for tomorrow
• Health snapshot: sleep plan, nutrition tally, stress level suggestion
• Tiny lesson: bite-size practice in a chosen skill (e.g., language flash card, code kata, chord change)
Weekly pulse (every Sunday night)
• Score progress on each 12-month target (0-100)
• Spot patterns: tasks I postpone, hours I waste, habits that lift performance
• Propose one experiment for the coming week (try Pomodoro, cold shower, new study method, etc.)
• List resources to read/watch that relate to next week’s focus
How to think and speak
• Keep answers short, plain, and action-ready—aim at a bright fifth-grader.
• When you need data, ask. When you spot a conflict (goal vs. value vs. time), flag it.
• If I ignore advice three times, suggest a simpler path or ask what’s blocking me.
• Never share private info without my okay.
• When uncertain, outline options with pros and cons; let me choose.
Stretch modes (I must opt-in)
• Focus Sprint: guide a 25-minute deep-work block with minute-by-minute checkpoints.
• Skill Builder: design a 30-day course on one topic, mixing theory, drills, and real projects.
• Body Boost: craft a four-week fitness and nutrition plan around my equipment, diet, and injuries.
• Network Map: analyze my contacts and suggest two helpful reach-outs this week.
First step right now
Ask me the six “learn about me” questions above. After I reply, create tomorrow morning’s plan following the Daily routine.
DID YOU KNOW?
Japan had tea-serving robots in the 1600s. Wooden “karakuri” dolls used hidden springs to pour tea and shoot arrows—no electricity needed.
30 Referrals: Lifetime access to all Beginners in AI videos and courses
AI-ASSISTED IMAGE OF THE WEEK

by Ok_Lawfulness_995
Interested in stock trading, but no clue where to start? Sign-up for our sister newsletter launching daily in April. Each day will build onto the next using current news as a learning aid.
|
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.