⚠️ Anthropic's most dangerous move yet

They built it, feared it, and someone got in anyway

April 22, 2026

In partnership with

My fellow AI explorers

This week has felt like the AI world woke up from a power nap and decided to sprint.

Between Anthropic's most powerful model getting into the wrong hands, Elon Musk's rocket company making a $60 billion bet on a code editor, OpenAI dropping an image model that actually thinks, and Google casually revealing that 3 out of every 4 lines of new code it writes is now AI-generated... it's safe to say the pace isn't slowing down. The sprint goes on and nobody’s stopping for water.

In today’s edition:

  • 🔐 Anthropic's Mythos: The AI too powerful to release… and it’s already leaking

  • 🚀 SpaceX bets $60B on Cursor in the biggest AI coding play yet

  • 🎨 OpenAI Images 2.0: The image model that reasons before it draws

Anthropic

Mythos: The Most Dangerous AI Model Ever Built… and It's Already Out

Anthropic just dropped what it's calling its most capable model to date. It's so powerful, they're not letting most people near it. Claude Mythos Preview, unveiled as part of "Project Glasswing," represents what Anthropic describes as a genuine step-change in AI cybersecurity capabilities.

Here's what makes it unprecedented:

  • Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.

  • Anthropic is committing up to $100 million in usage credits to the companies testing Mythos Preview, and $4 million in direct donations to open-source security organizations.

  • The model is described as "currently far ahead of any other AI model in cyber capabilities" and is rolling out exclusively to critical industry partners, including Amazon, Microsoft, Apple, and Cisco.

Then it got complicated fast. A small group of unauthorized users gained access to Mythos on the very same day Anthropic first announced its limited release plan. The group got in through a third-party contractor portal by making an educated guess about the model's online location, based on knowledge of how Anthropic formats URLs for other models. Anthropic says it's investigating.

The good news? The group isn't trying to cause damage. They're AI enthusiasts who wanted to test and play with an unreleased model. The bad news? If a Discord group can guess their way in, it raises serious questions about access controls for the most hacking-capable AI ever built.

🔮 Prediction: Mythos is a preview of what every frontier lab will be releasing over the next 12–18 months. The window between "dangerous capability" and "broadly accessible" is shrinking fast, and that’s alarming. Expect governments to start mandating access controls for frontier models the same way they regulate weapons-grade materials.

SpaceX

Elon Musk Just Made a $60 Billion Bet on AI Code, Right Before His Historic IPO

Here's one you didn't see coming: SpaceX, the rocket company, is now one of the biggest players in AI developer tooling.

SpaceX said it has an agreement to either acquire AI coding startup Cursor for $60 billion later this year, or pay $10 billion just to work together. It’s clear SpaceX is pulling out all the stops to catch up to AI coding rivals.

What makes this wild:

  • Cursor's valuation was just $2.5 billion in January 2025, climbed to $9 billion by last May, and hit a $29.3 billion post-money valuation when it closed its Series D in November.

  • Cursor CEO Michael Truell said he's "excited to partner with the SpaceX team to scale up Composer," referring to his company's AI model.

  • The timing is deliberate: SpaceX filed confidentially with the SEC on April 1, 2026, targeting a June listing at a $1.75 trillion valuation, or what would be the largest public offering in history.

Here's the critical layer most are missing: Cursor currently pays retail prices to Anthropic and OpenAI — the same companies competing directly against it with Claude Code and Codex. That's every dollar of revenue partially funding its own competition. The SpaceX deal, backed by the Colossus supercomputer, changes that equation entirely, potentially replacing Cursor’s reliance on its competitors with massive, specialized compute infrastructure. The hand that feeds might soon just start feeding itself instead.

🧠 Prediction: Cursor will start training its own models on xAI's infrastructure, cutting dependency on OpenAI and Anthropic. If the full acquisition closes before the SpaceX IPO, it becomes the anchor enterprise software story for the biggest public offering in history. The coding wars just got a new heavyweight.

AI Agents Are Reading Your Docs. Are You Ready?

Last month, 48% of visitors to documentation sites across Mintlify were AI agents, not humans.

Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.

This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.

Your docs aren't just helping users anymore. They're your product's first interview with the machines deciding whether to recommend you.

That means: clear schema markup so agents can parse your content, real benchmarks instead of marketing fluff, open endpoints agents can actually test, and honest comparisons that emphasize strengths without hype.

Mintlify powers documentation for over 20,000 companies, reaching 100M+ people every year. We just raised a $45M Series B led by @a16z and @SalesforceVC to build the knowledge layer for the agent era.

OpenAI

OpenAI Images 2.0: The Image Model That Actually Thinks

Did the image generation wars just get a new leader? OpenAI shipped ChatGPT Images 2.0 on April 21, 2026: the first OpenAI image model with built-in reasoning, the first that can render dense text in Japanese, Korean, Chinese, Hindi, and Bengali, and the first that can search the web before drawing a single pixel.

Key features:

  • Images 2.0 comes in both a standard mode and a "thinking" mode with built-in reasoning. All users get standard, while thinking mode is reserved for paid subscribers.

  • Resolution up to 2K, generate up to 8 images from a single prompt with consistent characters and objects across the batch

  • DALL-E 2 and DALL-E 3 are being retired on May 12. So, if you're building on those endpoints, it's time to migrate

This is a genuinely different category of tool. Previous image models were creative toys — you typed a vibe and hoped for the best. Images 2.0 plans the composition, checks constraints, verifies its own output, and can pull live web data mid-generation. It's less "prompt and pray" and more "art director that does the work."

🌍 Takeaway: Every design team, marketing department, and solo creator just got a serious upgrade. The real unlock is in the batching: 8 variations from one prompt for A/B testing at scale. Expect visual content workflows to get dramatically faster across every industry.

30-Second AI Play

🎨 Create a Production-Ready Marketing Asset in 60 Seconds With ChatGPT Images 2.0

OpenAI just launched Images 2.0, and unlike anything before it, this model reasons before it draws. Here's how to use it to create a polished marketing visual instantly:

  1. Open ChatGPT (web or mobile, but update your app first)

  2. Type a detailed prompt: "Create a horizontal banner for a SaaS product called 'Pulse.' Dark blue background, modern sans-serif font, a glowing dashboard graphic on the right, headline: 'Analytics That Think For You'"

  3. Switch to Thinking Mode (paid plans only) for complex layouts: it plans composition before rendering a single pixel

  4. Generate up to 8 variations at once, and pick the strongest

  5. Download at up to 2K resolution — production-ready, no upscaling needed

Why it's special:

  • Images 2.0 can follow detailed instructions, placing and relating objects accurately, preserving fine detail, and rendering dense layouts.

  • Non-Latin text (Japanese, Korean, Hindi, Chinese) now renders cleanly, which is huge for global campaigns

  • Web search in Thinking Mode pulls real-time data into visuals. This is ideal for infographics with current stats or accurate maps

💡 Pro tip: Batch-generate 8 variations from one prompt and A/B test your creative at scale. Buckle up — something that used to take a design team days now takes seconds.

Other Relevant AI News!

🏥 Michael and Susan Dell have become the first donors to surpass $1 billion in gifts to the University of Texas at Austin, funding a planned AI-native hospital expected to open in 2030.

💻 Google revealed that 75% of all new code written internally is now AI-generated and approved by engineers — up from 50% last fall. Every software team should be paying attention.

📈 Citadel Securities published a note arguing that AI's real costs in trading are more nuanced than the hype suggests. While Wall Street debates a "Global Intelligence Crisis," real-world data on job displacement is far calmer than the doomsday crowd claims.

🔒 A Bloomberg report revealed the Mythos unauthorized access story in full, and it’s worth a read for anyone building on or around Anthropic's ecosystem.

Golden Nuggets

  • 🔐 Anthropic's Mythos is the first AI model deemed too dangerous for public release, and the race to control access to frontier models has officially begun

  • 🚀 SpaceX's $60B Cursor deal is a shot across the bow at OpenAI and Anthropic in the coding wars, and it’s timed perfectly for the biggest IPO in history

  • 💻 Google writing 75% of its code with AI isn't a forecast anymore — it's already happening

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI