OpenAI Raises $122B, Claude Code Leaks Its Secrets | Weekly Digest
$122B nobody wants, 512,000 lines nobody was supposed to see, and the AI industry's strangest week yet
Hey! Welcome to the latest Creators’ AI Edition.
OpenAI raised more money in a single round than the GDP of Hungary. Anthropic accidentally shipped the source code of its most valuable product to the entire internet. And SpaceX filed for what could be the largest IPO in human history — with Grok inside. Welcome to the last week of Q1 2026. Let’s break it down.
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
Keep your mailbox updated with practical knowledge & key news from the AI industry!
Featured Materials 🎟️
OpenAI Raises $122 Billion at $852 Billion Valuation 💰
On March 31, OpenAI closed the largest private funding round in history — $122 billion at an $852 billion post-money valuation. The round was anchored by three major investors: Amazon ($50 billion, of which $35B is contingent on an IPO or reaching AGI), NVIDIA ($30 billion), and SoftBank ($30 billion). For the first time, OpenAI also raised $3 billion from individual investors through bank channels and will be included in ARK Invest ETFs.
The numbers behind the raise:
$2B/month in revenue, growing 4x faster than Alphabet and Meta at the same stage
900 million weekly active ChatGPT users
50M+ paid subscribers
15 billion tokens processed per minute via API
Codex now serves 2M weekly users — up 5x in three months
Ads pilot hit $100M ARR in under six weeks
Enterprise is 40% of revenue, on track for parity with consumer by end of 2026
What it’s for:
OpenAI is building what it calls a “unified AI superapp” — combining ChatGPT, Codex, browsing, and agentic capabilities into one product. The company has also discontinued the standalone Sora app, redirecting compute toward its higher-revenue priorities.
The twist:
The day after the record raise, Bloomberg reported that $600 million in OpenAI shares had been sitting unsold on secondary markets — six institutional investors couldn’t find a single buyer. Meanwhile, across secondary platforms, $2 billion in cash was waiting to deploy into Anthropic. OpenAI’s enterprise API market share dropped from 50% to 25% over the past year, while Anthropic’s rose from 12% to 32%. OpenAI reportedly labeled this gap a “red alert” internally. The largest fundraise in history and the quietest secondary market in years — happening simultaneously.
The bigger picture: This round is as much about anchoring IPO expectations as it is about the capital. Both OpenAI and Anthropic are targeting public listings in Q4 2026.
Source: Bloomberg | CNBC | TechCrunch | OpenAI Blog
Claude Code Accidentally Leaked Its Entire Source Code 🔓
On March 31 at 04:23 UTC, Anthropic published Claude Code v2.1.88 to npm — and accidentally shipped .map sourcemap files that fully reconstructed nearly 2,000 TypeScript files and 512,000 lines of code. No exploit, no credentials, no special access required. A security researcher flagged it publicly within hours. The post accumulated 34 million views. A mirrored copy appeared on GitHub and crossed 84,000 stars. Anthropic pulled the version the same day and used copyright takedown requests to remove 8,000+ copies.
What the leak revealed:
Undercover Mode — Claude Code is instructed via system prompt: “You are operating UNDERCOVER in a PUBLIC/OPEN-SOURCE repository. Your commit messages, PR titles, and PR bodies MUST NOT contain ANY Anthropic-internal information. Do not blow your cover.” In other words, when developers use Claude Code to contribute to public repos, the tool actively conceals its involvement.
Anti-distillation poisoning — controls that inject fake tool definitions into API responses if competitors attempt to scrape Claude Code outputs to train their own models.
Three-layer memory architecture — the solution to “context entropy” that Anthropic spent years building. A lightweight index (MEMORY.md), topic files fetched on-demand, and a grep-only transcript layer. Competitors now have the blueprint.
KAIROS / autoDream — a background daemon mode where Claude Code performs “memory consolidation” while the user is idle — merging observations, resolving contradictions, converting insights into facts.
Claude Mythos (codename: Capybara) — multiple references to an unreleased model described in a separate leaked document as a “step change in capabilities,” with a significantly larger context window than anything currently on the market.
Frustration tracking — code that scans user prompts for profanity, insults, and phrases like “this sucks,” and appears to log expressions of negativity.
The human error behind the leak:
Anthropic acquired the Bun JavaScript runtime at the end of 2025. A known Bun bug (issue #28001, filed March 11) caused sourcemaps to ship in production builds. The bug was open for 20 days before the leak. Nobody caught it.
For creators and builders: The most commercially significant finding isn’t the “Undercover Mode” — it’s the three-layer memory architecture. If you’re building AI agents that need to maintain coherent context over long sessions, this leaked blueprint is the most detailed public documentation of a production solution available anywhere.
SpaceX Files for the Largest IPO in History 🚀
SpaceX has confidentially filed for an IPO targeting a $1.75 trillion valuation and a $75 billion raise — which would exceed the GDP of more than half the world’s countries and shatter Saudi Aramco’s previous record of $29.4 billion. This is not just a rocket company going public. In February, SpaceX absorbed xAI in an all-stock deal at a $1.25 trillion combined valuation. It also owns X (formerly Twitter). The entity going public spans orbital launches, satellite internet (Starlink), AI (Grok), and social media — all controlled by one individual.
The numbers:
Starlink: $10.6B revenue in 2025, 54% EBITDA margins, 10M+ subscribers across 150 countries
SpaceX total 2025 revenue: $16B
xAI component: reportedly burning ~$1B/month
All 11 original xAI co-founders have departed since the merger
The AI angle:
Musk said in March that xAI “was not built right the first time around, so is being rebuilt from the foundations up.” SpaceX recently hired two engineers from Cursor to bolster its AI coding efforts. The IPO would be the first major public test of whether xAI’s Grok can be valued alongside Starlink’s cash flow — or whether it drags the valuation down.
SpaceX, OpenAI, and Anthropic together could raise more than $150 billion in IPOs by the end of 2026 — roughly as much as all US VC-backed company IPOs combined over the past decade. PitchBook warns this concentration of capital may crowd out smaller listings entirely.
Source: The Next Web | BusinessToday | Fortune
News of the week 🌍
DeepSeek Goes Dark for 7 Hours — Right Before V4 🔻— DeepSeek experienced its longest outage since going viral in early 2025 — 7 hours and 13 minutes of full service disruption, affecting users and API-dependent products globally. No official cause was given. The timing is notable: V4, DeepSeek's next major model, is expected in April, and industry observers speculate the downtime reflects infrastructure work ahead of the launch. For teams evaluating DeepSeek as a low-cost alternative to OpenAI and Anthropic APIs, this is a concrete reliability signal — good model, still maturing infrastructure.
Google Search Live Launches in 200+ Countries 🌐 — Google’s real-time multimodal search — point your phone at anything, ask a question out loud, get an AI answer live — rolled out globally, powered by Gemini 3.1 Flash Live. Available across all languages where AI Mode is active. Also released as an API in Google AI Studio. Real-time voice+camera AI search is now a global-default product, not a US experiment.
Mistral Raises $830M in Debt for European AI Infrastructure 🇪🇺 — France’s Mistral secured $830M from a seven-bank group to buy 13,800 NVIDIA chips and build a major data center near Paris. Europe’s clearest move yet toward AI sovereignty — owned compute, local infrastructure, a target of 200 megawatts across Europe by end of 2027.
The Same Morning as the Claude Code Leak, npm Got Hit by a RAT ☠️ — At 00:21 UTC on March 31 — four hours before Anthropic accidentally shipped their source code to npm — attackers planted malicious versions of the axios package (1.14.1 and 0.30.4) in the same registry, carrying an embedded Remote Access Trojan. Two separate npm security incidents in one morning: one accidental, one deliberate. Any developer who ran npm install on March 31 had two separate reasons to audit their lockfiles immediately.
Runway Launches $10M Fund for AI Video Startups 🎬 — Runway is seeding early-stage startups building on its video models — free API credits, a Builders program, and direct investment checks. The AI video company is positioning itself as ecosystem infrastructure rather than just a destination product. For founders building video-native AI applications, this is the most accessible compute funding available right now.
Anthropic Research: Claude Has Internal Emotion Representations 🧠 — Anthropic published research finding that Claude has internal representations of “emotion concepts” that actually drive its behavior — sometimes in surprising ways. The paper raises new questions about what’s actually happening inside frontier AI systems at inference time.
Useful tools ⚒️
⭐ Jupid — File your taxes with AI. Connects to your financial accounts, uses Claude to analyze transactions and fill your tax return end-to-end.
Notion MCP — Your Notion workspace inside every AI agent. Lets any MCP-compatible AI agent read, write, and query Notion databases directly — no manual copy-paste between tools.
Computer Use in Claude Code — Let Claude use your computer from the CLI. Native computer use built directly into the terminal workflow — click, type, scroll, interact with any app.
Ollama v0.19 — Massive local model speedup on Apple Silicon with MLX. If you run models locally on a Mac, this update changes the experience entirely. Decode speed nearly doubles, prefill 1.6x faster. Requires 32GB+ unified memory — check before updating.
Pixero AI — Autonomous AI agents inside your ad campaigns — creative generation, targeting, A/B testing, all agent-native. No manual setup required.
Know friends, co-founders or vibe-coders who need this content. Share!
Weekly Guides 📕
Gemini Live API — Official Developer Documentation — Google's official guide to building real-time voice+vision agents on Gemini 3.1 Flash Live: WebSocket sessions, multimodal input handling, function calling, and ephemeral token setup.
Ollama 0.19 + MLX on Apple Silicon — Official Setup — Ollama's own walkthrough of the v0.19 MLX backend: what changed under the hood, which models are supported, benchmark numbers (1.6x prefill, 2x decode), and the 32GB memory floor to check before updating.
Migrating from Gemini 2.5 Flash Live to 3.1 — Google's official migration checklist: new model string, changed thinking config, and the three features — proactive audio, affective dialog, async tool calling — that do not carry over from 2.5. Required reading before you touch an existing voice agent.
Install Claude Code Without npm — Native Installer — Anthropic's official response to the March 31 supply chain incident: a standalone binary that bypasses npm's dependency chain entirely. One curl command, no lockfile to audit.
Claude Code Architecture Patterns from the Leak — VentureBeat Deep Dive — How to implement the three-layer memory architecture, anti-distillation traps, and multi-agent orchestration patterns revealed in the leaked source — the closest thing to official agent architecture docs that exists right now.
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
DeepSeek's Worst Outage Ever — What It Tells You About AI Infrastructure — Why 7 hours of downtime reveals more about DeepSeek's trajectory than any benchmark: the gap between building a capable model and running production infrastructure at the scale of hundreds of millions of users.
Anthropic Research: Emotion Concepts in Claude — What Anthropic’s own researchers found when they looked for emotional representations inside the model
The Axios npm Supply Chain Attack: Full Technical Breakdown — Security Boulevard— The complete account of the malicious axios 1.14.1/0.30.4 packages planted by North Korea-nexus threat actor UNC1069 — how the RAT worked, what to check in your lockfiles, and why this happened on the same morning as the Claude Code leak.
April 2026 AI Model Landscape: Everything That’s Coming — Claude Mythos, GPT-5.5 (Spud), DeepSeek V4, Grok 5 — the full Q2 model pipeline
If you missed our previous updates, don’t worry, here they are:
Claude Mythos Leaked, ARC-AGI-3 Humiliates Every AI, Sora Is Dead