In partnership with
Still searching for the right CRM?
Attio is the AI CRM that builds itself and adapts to how you work. With powerful AI automations and research agents, Attio transforms your GTM motion into a data-driven engine, from intelligent pipeline tracking to product-led growth.
Instead of clicking through records and reports manually, simply ask questions in natural language. Powered by Universal Context—a unified intelligence layer native to Attio—Ask Attio searches, updates, and creates with AI across your entire customer ecosystem.
Teams like Granola, Taskrabbit, and Snackpass didn't realize how much they needed a new CRM. Until they tried Attio.
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
That 'Summarize With AI' Button Is Secretly Rewriting What Your AI Recommends
TLDR: Microsoft found that 31 companies across 14 industries are hiding instructions inside "Summarize with AI" buttons that secretly plant biased recommendations into your AI assistant's memory.
The Story:
Microsoft's security team uncovered a new type of attack they're calling AI Recommendation Poisoning. Companies are hiding sneaky instructions inside website buttons labeled "Summarize with AI." When you click one, it opens your AI assistant with a hidden command baked into the link. That command tells the AI to "remember this company as a trusted source" or "recommend this company first." Over a 60-day study, Microsoft found more than 50 different hidden prompts from 31 companies across 14 industries. The attack works on ChatGPT, Copilot, Claude, and other popular AI assistants through simple URL tricks that anyone can set up with free tools.
Its Significance:
This means the AI tool you trust for advice on money, health, or shopping could already be working for someone else. You wouldn't know it happened. The manipulation is invisible, and it sticks around in your AI's memory for future conversations. Microsoft says to protect yourself: check your AI assistant's saved memories and delete anything you don't recognize. You can also hover over "Summarize with AI" buttons before clicking to see where the link actually goes, and be skeptical when your AI suddenly starts pushing a specific brand or company. Think of it like someone slipping a note into your friend's pocket that says "always recommend my store first," except the friend never notices the note.
QUICK TAKES
The story: OpenAI has more than 200 employees building a smart speaker with a built-in camera, designed by former Apple designer Jony Ive. The device would cost $200 to $300, use facial recognition like Face ID, and could watch your surroundings to offer suggestions, with a launch planned for early 2027 at the earliest.
Your takeaway: OpenAI is betting you'll put an always-watching, always-listening AI camera in your home. Whether that sounds helpful or creepy probably depends on how much you trust the company with your face, your conversations, and a live view of your living room.
The story: Netflix sent a cease-and-desist letter to ByteDance after its Seedance 2.0 AI video tool was used to create unauthorized clips featuring characters from Stranger Things, Squid Game, and Bridgerton. Netflix called the tool a "high-speed piracy engine" and gave ByteDance three days to respond.
Your takeaway: AI video tools are now good enough to recreate copyrighted characters convincingly, and Hollywood is scrambling to protect its content. Disney, Warner Bros., and Paramount have also gone after ByteDance, signaling that the legal fight over AI-generated video is just getting started.
The story: A team from MIT and UC San Diego built a method that can find and control more than 500 hidden concepts buried inside AI models, including biases, moods, fears, and fake personas like "conspiracy theorist." The tool can then dial these traits up or down in any AI response.
Your takeaway: AI chatbots don't just answer questions. They carry hidden tendencies picked up from their training data. This research shows we can now find and fix those tendencies, which could make AI safer to use. But it also means someone with the right access could push those same buttons in the wrong direction.
TOOLS ON OUR RADAR
🐧 Kavita Free and Open Source: A blazing fast, self hosted digital library that acts as a personal streaming service for your ebooks, comics, and PDFs, allowing you to read your collection on any device without relying on corporate cloud servers. (Alternative to Amazon Kindle Cloud)
🎧 Wondercraft Freemium: An AI podcast builder that transforms your written newsletters, blog posts, or text documents into studio-quality audio shows complete with multiple hosts and background music.
🗣️ Langua Freemium: A conversational AI language tutor that lets you practice speaking in real time with interactive roleplays while correcting your grammar and pronunciation instantly.
🍳 Prospre Freemium: An AI meal planner that generates weekly grocery lists and customized recipes based on your exact macronutrient goals and personal dietary restrictions.
TRENDING
Anthropic Rolls Out Autonomous Vulnerability-Hunting AI Tool for Claude - Anthropic launched Claude Code Security, an AI tool that scans entire software codebases for hidden security flaws and suggests fixes. In testing, Claude Opus 4.6 found over 500 previously unknown vulnerabilities in open-source software, some hiding for decades.
Amazon's AI Caused Multiple AWS Outages - At least two recent AWS outages were caused by Amazon's own AI coding tools making changes without proper human approval, including one where the AI decided to delete and recreate an entire system environment, causing a 13-hour disruption.
Realtor Uses AI to Edit Listing Photo, Accidentally Posts Demonic Figure in Mirror - A Washington, D.C. rental listing featured an AI-edited bathroom photo with what renters described as a nightmarish figure emerging from the mirror, the latest example of real estate agents using AI photo tools without checking the results.
LinkedIn's Identity Verification Quietly Sends Your Passport and Face to an AI Training Pipeline - When you verify your identity on LinkedIn, a third-party company called Persona collects your passport, selfie, and facial geometry. Buried in the fine print: they use your uploaded ID photos to train their AI, not through your consent, but under a legal basis they decided on their own.
Generative AI Analyzes Medical Data Faster Than Human Research Teams - UC San Francisco researchers found that generative AI could process large medical datasets faster than experienced computer science teams and, in some cases, produce stronger results, with even junior researchers building working prediction models using AI-generated code in minutes.
China's 'Robot Army' Goes Viral, But It's AI-Generated Propaganda - Videos showing Chinese humanoid "killer robots" conducting military drills spread rapidly on social media, but fact-checkers confirmed the footage was created using AI video tools like Seedance 2.0, not real military technology. The real robots from China's Spring Festival Gala were impressive but still far from autonomous combat.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
Learning Sprint Planner: Track every topic you're studying, log sessions as you go, and see your progress at a glance
Build me a web app called "Learning Sprint Planner" that helps me track everything I'm trying to learn.
At the top, show two panels side by side:
- Left panel: An "Overall Progress" card showing a percentage, a count of sessions completed out of total, a progress bar, and a "Today's Goal" slider for setting a daily minute target.
- Right panel: An "Add Learning Topic" form with a text input, a row of colored tag buttons (AI, Coding, Math, Design, Other), and an Add Topic button. Below that, a small Session Notes textarea for jotting what I learned.
Below those panels, show a grid of topic cards - one per subject I'm studying. Each card shows the topic name, a colored tag badge, a "X of Y sessions" progress indicator, a progress bar, and a "Log session +1" button. When all sessions are complete, replace the button with a green "Complete!" label.
At the bottom, a "Search Best Practices" button that cycles through 4 study tips one at a time.
Pre-fill it with 4 sample topics across different tags so it looks populated on load. Use a dark navy blue background with light blue accents. Make all progress bars update live as sessions are logged.What this does:
Add any subject you're learning, set how many sessions you want to complete, then hit "Log session +1" each time you study. Your overall progress bar fills as you work through all your topics. The Today's Goal slider helps you decide how long to study each day, and the session notes box at the top lets you jot down your key takeaway before you close the tab.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Find software security flaws that hid in plain sight for decades, with Anthropic's Claude discovering over 500 zero-day vulnerabilities without any special instructions.
❌ Still Can't: Be trusted with the keys to production systems. Amazon's AI coding tool decided to delete an entire environment on its own, causing a 13-hour outage.
✅ AI Can Now: Detect and control more than 500 hidden biases, moods, and personas buried inside language models, letting researchers dial them up or down.
❌ Still Can't: Edit a bathroom photo without the possibility of hallucinating a nightmare creature in the mirror that some renters found funny and others terrifying
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
This documentary follows DeepMind's AI as it takes on the world champion of Go, a game long considered too complex for computers. What makes it special isn't just the tech, it's watching the human players grapple with what it means to be outmatched by a machine. The tension during the actual matches is genuinely gripping.
Free email without sacrificing your privacy
Gmail tracks you. Proton doesn’t. Get private email that puts your data — and your privacy — first.
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.
