🚨 Anthropic Leaks Claude Code source code

Good Morning! Anthropic is having a catastrophic week for operational security, accidentally leaking the entire source code for its flagship AI coding tool just days after exposing its most powerful unreleased model. Plus I’ll show you how to turn simple chats into permanent, reusable agent apps.

Plus, in today’s AI newsletter:

• Anthropic Accidentally Leaks Claude Code Source Code

• Google’s Quantum AI Puts a Ticking Clock on Crypto

• Google Doubles Down on Video with Veo 3.1 Lite

• How to Turn Simple Chats Into Permanent, Reusable Agent Apps

• 4 new AI tools worth trying

AI TOOLS

🚨 Anthropic Accidentally Leaks Claude Code Source Code

Anthropic has inadvertently exposed over 512,000 lines of unobfuscated TypeScript source code for Claude Code, its massively popular agentic CLI tool. The leak happened after a debugging .map file was accidentally bundled into a public npm release, giving developers a complete blueprint of the software's inner workings.

• The leaked code reveals the "agentic harness", the complex system of permissions, tool-call loops, and guardrails that instructs the underlying AI how to operate autonomously on a user's machine.

• Anthropic confirmed the leak was a "packaging issue caused by human error" rather than a hack, and stated that no underlying model weights or customer data were compromised.

• The codebase was quickly mirrored across GitHub, potentially allowing competitors or open-source developers to reverse-engineer Anthropic's highly lucrative enterprise tool.

• This massive blunder comes just days after Anthropic accidentally left nearly 3,000 internal documents public on its CMS, revealing an unreleased, ultra-powerful model tier internally dubbed "Mythos" (and codenamed "Capybara").

Anthropic has built its entire reputation on safety and security, but back-to-back operational blunders are damaging that image. While no AI model weights were stolen, leaking the proprietary orchestration layer for Claude Code gives rivals a literal blueprint for building high-agency, commercially viable AI agents. Meanwhile, the "Mythos" leak confirms that the next generation of frontier models will possess cybersecurity capabilities so advanced that they require entirely new, heavily restricted deployment strategies.

QUANTUM AI

🔐 Google’s Quantum AI Puts a Ticking Clock on Crypto

Google Quantum AI has released a white paper demonstrating that the elliptic curve cryptography securing most major cryptocurrencies is far more vulnerable to quantum attacks than earlier estimates suggested.

• Improved quantum algorithms mean breaking 256-bit encryption now requires roughly 1,200 logical qubits and under 500,000 physical qubits, a massive reduction in the required hardware scale.

• Advanced quantum systems could theoretically execute an "on-spend" attack (hijacking a live transaction before it is confirmed on the blockchain) in a matter of minutes.

• Dormant digital assets are particularly vulnerable to "at-rest" attacks, as wallets with exposed public keys cannot be actively upgraded to new post-quantum cryptographic standards.

• In a push for responsible disclosure, Google withheld the actual attack circuits, instead publishing a zero-knowledge proof to validate their mathematical claims without arming bad actors.

• While hardware capable of this doesn't exist just yet, the rapidly shrinking resource gap increases the urgency for the crypto industry to begin transitioning to post-quantum cryptography (PQC) immediately.

The buffer zone between current quantum hardware and the ability to crack Bitcoin and Ethereum is evaporating much faster than anticipated. By proving that the mathematical threshold to break blockchain encryption is significantly lower, Google is forcing a massive, decentralized industry to face an existential threat. Upgrading an entire ecosystem to post-quantum standards takes years of coordination, and the clock is officially ticking.

AI MODELS

🎥 Google Doubles Down on Video with Veo 3.1 Lite

Following OpenAI's sudden exit from the AI video race last week, Google announced its unwavering commitment to the medium by launching Veo 3.1 Lite, its most cost-effective video generation model to date.

• Designed for high-volume applications, Veo 3.1 Lite slots beneath Veo 3.1 Fast, offering the exact same generation speeds but at less than half the cost.

• The model supports both Text-to-Video and Image-to-Video workflows in 720p and 1080p resolutions, covering both landscape (16:9) and portrait (9:16) aspect ratios.

• Developers can now tightly control their spend by customizing video durations to 4, 6, or 8 seconds, with pricing scaling dynamically based on the length.

• The new model is rolling out today on the Gemini API and Google AI Studio, alongside news that the higher-tier Veo 3.1 Fast model will receive a price cut on April 7.

With OpenAI stepping back from Sora, Google is aggressively seizing the opportunity to corner the AI video market. By releasing a cheaper, highly customizable model that integrates directly into enterprise APIs and its massive consumer ecosystem, Google is signaling that AI video isn't just a flashy research experiment, it's a scalable, foundational product ready for mass deployment.

HOW TO AI WITH CREAO

🗂️ How to Turn Simple Chats Into Permanent, Reusable Agent Apps

There is a new AI that lets you describe any app or workflow in plain language and instantly builds a fully functional, AI-native tool you can use, share, and reuse, no coding experience needed.

🧰 Who is This For

• Business owners who need custom internal tools without hiring developers

• Founders who want competitor monitoring on autopilot

• Sales teams automating lead research across multiple sources

• Anyone tired of forcing their work into one-size-fits-all SaaS tools

STEP 1: Create Your Free Account and Open a Workspace

Head over to CREAO's website and sign up for a free account. No credit card is required. Once you're in, you'll land on your main dashboard. From here, click "New Workspace" to get started. Think of the workspace as your project home, it's where your Super Agent, apps, and integrations all live together and compound over time.

STEP 2: Describe Your App to the Super Agent

In plain language, tell the Super Agent exactly what you want to build. Be specific about the problem it solves and how you'll interact with it.

For example:

"Monitor 3 competitor websites every Monday, log pricing changes to Google Sheets, and send me a Slack alert if anything moves more than 10%."

The Super Agent will process your request and present a detailed plan before building anything. Review it carefully, if something doesn't match your vision, just type your feedback in the chat and it will adjust.

STEP 3: Connect Your Integrations

Once your app plan is confirmed, the Build Agent will ask you to select the integrations your app needs. It supports a wide range of tools including Slack, Gmail, Google Workspace, Notion, and custom APIs.

Only select the integrations you actually need. This helps the agent stay focused and deliver better results. For the competitor monitoring example, you’d connect Google Sheets to log the pricing data and Slack to receive the automated alerts whenever a price moves beyond your threshold.

STEP 4: Test, Refine, and Save as a Reusable Agent App

Run it once to verify the output. If the Sheets formatting is wrong, describe the fix in chat. When it works, save it as an Agent App, it runs on schedule automatically from that point on.

Set it to run daily, weekly, or on any schedule you choose. Your Agent App executes deterministically, no AI involved after the first setup. From that point on, your team gets reliable, consistent output without touching a thing.

Pro Tip: Start simple. The most powerful setups are built by layering small, focused apps over time, not by trying to build everything at once.

OpenAI closed a $122B funding round led by SoftBank, a16z, and others at an $852B post-money valuation, after previously announcing the round would total $110B.

Oracle is cutting thousands of jobs in its latest layoffs as the company continues to ramp AI spending; as of May 2025, Oracle employed 162,000 people.

Google says all users in the US can now change their Google Account username; users are restricted to one username change every 12 months.

Google plans to release a screenless Fitbit band later this year; it will include basic features and require a paid subscription for more functionality.

🎶 Lyria 3 Pro: Google’s AI that makes longer music tracks

🧠 Qwen3.5-Omni: Alibaba’s AI that understands text, images, audio, and video

🤖 Hermes Agent: AI agent with memory that works across different platforms

🔎 Model Council: Perplexity tool to query multiple AI models at once

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas