AI Weekly · April 24, 2026

🤖 AI Weekly: Anthropic's secret AI found a 27-year-old security...

Anthropic's secret AI found a 27-year-old security flaw hiding in plain sight | ChatGPT's new image tool is a huge leap — and it can finally spell |...

← All Issues

Your weekly briefing on the AI stories, trends, and tips that matter most. Curated for the curious, not just the technical.

Big Stories This Week

Model Release

Anthropic's secret AI found a 27-year-old security flaw hiding in plain sight

7 sources · 33 developments · Continuing coverage

Anthropic built a powerful AI called Mythos and kept it out of public hands, using it instead to hunt for security holes in widely-used software. It found a 27-year-old weakness in a system considered one of the most secure in the world, along with thousands of other flaws across major browsers and operating systems.

Why it matters

The apps and devices you use every day are quietly getting safer because an AI is finding and helping fix problems humans missed for decades.

Model Release

ChatGPT's new image tool is a huge leap — and it can finally spell

2 sources · 4 developments

OpenAI released GPT Image 2.0 this week, and it blew past every other image generator by the widest margin anyone has tracked. It can now write readable text inside pictures in languages like Japanese, Hindi, and Chinese, and even build believable 360-degree panoramas.

Why it matters

If you've ever tried to make an AI picture with words on it — a birthday card, a menu, a poster — this is the first version that actually gets the letters right.

Product Launch

Anthropic launches Claude Design, a new tool for making prototypes and slides

11 sources · 14 developments

Anthropic rolled out Claude Design, a web-based tool that lets you create wireframes, slideshows, and short videos by describing what you want. It joins a growing suite of Claude tools aimed at helping regular people build things without needing design skills.

Why it matters

If you've wanted to mock up a website idea, a pitch deck, or a flyer but don't know Photoshop or Figma, this gives you a free-feeling way to get it done just by typing.

Things to Try This Week

1. Ask AI to compare two things side by side

If you're trying to decide between two options — like two products, two vacation spots, or two recipes — AI can lay out the differences for you in seconds. It's like having a friend who has done all the research already.

  1. Open ChatGPT, Claude, or Gemini (they're all free at chatgpt.com, claude.ai, or gemini.google.com).
  2. Type this exact thing: 'Compare a Toyota RAV4 and a Honda CR-V for a family with two kids. List the pros and cons of each in simple terms.'
  3. Read what comes back, then try swapping in any two things you're deciding between.

Source: Ranking Every AI Tool I Used in 2026 (What's ACTAULLY Good)

2. Have AI plan your day for you

If you've got a long to-do list and don't know where to start, AI can look at everything and tell you what to tackle first. It's like having a patient friend help you think through your priorities.

  1. Open ChatGPT, Claude, or Gemini in your browser.
  2. Type this exact thing: 'Here's my to-do list for today: reply to 10 emails, write a birthday card for my mom, finish a work report, go grocery shopping, and call the dentist. What order should I do these in, and why?'
  3. Read the suggestion and tell it about anything it got wrong — it will adjust.

Source: I Replaced OpenClaw and Hermes With This Claude Code Setup

3. Ask AI to look at a picture and tell you about it

You can upload a photo to free AI tools and ask questions about what's in it — a plant you can't identify, a strange food label, a handwritten note you can't read. It's like having a knowledgeable friend available any time.

  1. Open ChatGPT or Gemini on your phone or computer (both free).
  2. Click the little paperclip or photo icon, pick a picture from your phone (try a plant in your house), and type: 'What kind of plant is this and how often should I water it?'
  3. Read the answer, then try it again with a different photo — maybe a food label or a receipt.

Source: Opus 4.7 Is GREAT (except the token usage)