In partnership with
In Today’s Issue:
📉 Anthropic accidentally leaked the entire source code of Claude Code
🤑 OpenAI reportedly closed a jaw-dropping $122 billion mega-round
🩺 A New York hospital CEO suggested AI could soon replace radiologists
🌞 Solar power is leading the charge as global renewable energy capacity surges
✨ And more AI goodness…
Dear Readers,
Anthropic just accidentally shipped its entire playbook: 512,000 lines of Claude Code source, unreleased features, secret model codenames, and all, buried in a sourcemap file anyone could download from npm. It's the kind of leak that makes you rethink what "move fast and break things" means when you're building the world's most popular AI coding agent, and it's our featured story today.
But that's far from all. OpenAI closed a jaw-dropping $122 billion mega-round that rewrites the rules of venture funding, a New York hospital CEO is openly saying AI could replace radiologists (cue the firestorm), renewables just hit 49.4% of global electricity capacity with solar doing most of the heavy lifting, and, perhaps most quietly stunning, an unreleased OpenAI model produced three publishable proofs to open Erdős problems in mathematics entirely on its own. No human guidance needed.
Grab your coffee, this one's packed.
All the best,
Kim Isenberg
🚀 OpenAI Lands Record Mega-Round
OpenAI has reportedly closed a staggering $122 billion funding round, described here as the largest in Silicon Valley history, with backing from major names like Amazon, Nvidia, SoftBank, plus wealthy investors. The move matters because it not only supercharges OpenAI ahead of an expected IPO by the end of 2026, but also expands access for individual investors through more than $3 billion raised via banks and planned inclusion in ARK Invest ETFs.
🩺 AI Sparks Radiology Replacement Debate
NYC Health + Hospitals CEO Mitchell Katz says AI could already take over a large share of radiology work, especially first reads for mammograms and X-rays, once regulations allow it. Supporters argue this could expand screening access and cut costs, with one hospital leader claiming their breast-screening AI is wrong only about 3 times in 10,000 negative cases. Radiologists are pushing back hard, however, warning that AI-only reads could put patients at serious risk.
🌞 Renewables Near Half Global Capacity
Renewables surged to 49.4% of global electricity capacity in 2025, hitting a record 5,149 GW after adding 692 GW in one year, with solar alone contributing a massive 511 GW. That pushed annual renewable growth to 15.5%, bringing the world very close to the COP28 goal of tripling renewable capacity by 2030.
The big takeaway is that solar is now the main engine of the energy transition, while renewables are also looking more attractive from an energy security standpoint as fossil fuel markets stay vulnerable to geopolitical shocks. Wind also kept expanding strongly, and the broader trend suggests clean power is becoming both a climate solution and a resilience play for countries worldwide.
Jensen Huang explains why TPU can't beat Nvidia. Bold take, competition is heating up.
Claude Code's Source Exposed
The Takeaway
👉 Anthropic accidentally published Claude Code's full source — roughly 512,000 lines of TypeScript — via a sourcemap file in an npm package, exposing unreleased features, internal model codenames, and the complete tool architecture behind the most popular AI coding agent.
👉 The codebase reveals a product far ahead of its public release: an always-on assistant mode (KAIROS), a background memory "dream" system, multi-agent orchestration, and references to unreleased models including one codenamed "Capybara" — suggesting Anthropic's shipping cadence is deliberately gated, not capability-limited.
👉 This is Anthropic's second data exposure in under a week, following a CMS misconfiguration that revealed details about an unreleased model — a pattern that raises questions about internal release processes at a company now generating billions in revenue.
👉 The root cause was a build configuration oversight: Bun's bundler generates sourcemaps by default, and no step excluded the file before publishing to npm — a reminder that even the most sophisticated engineering teams can be undone by a missing line in .npmignore.
Anthropic just handed the world its playbook, accidentally. It is probably the biggest leak of the year. A sourcemap file bundled into Claude Code's npm package exposed roughly 512,000 lines of TypeScript, revealing the full internal architecture of the most popular AI coding agent on the market. Security researcher Chaofan Shou spotted the exposure and posted the discovery on X, where it quickly amassed millions of views.
What's inside is remarkable. The leaked code reveals unreleased features like KAIROS, an always-on background assistant that "dreams," running memory consolidation while you're idle. There's a full multi-agent orchestration system, a Tamagotchi-style companion pet called "Buddy," and references to unreleased models including one codenamed "Capybara." An ironic highlight: Anthropic built an entire "Undercover Mode" to prevent internal information from leaking through AI-generated commits, then shipped the whole source in a .map file.
Anthropic confirmed the incident was caused by human error in the release packaging process, not a security breach, and stated that no customer data was involved. This marks the second accidental data exposure from Anthropic in under a week, following a CMS misconfiguration that revealed details about an unreleased model called "Claude Mythos."
The engineering underneath, though, is genuinely impressive; this isn't a simple API wrapper but a deeply considered system with 40+ tools, sophisticated permission layers, and multi-agent coordination. For anyone building in the AI tooling space, this is an unprecedented look at what production-grade agentic infrastructure actually looks like.
Why it matters: This leak gives developers and competitors an unfiltered blueprint of how the leading AI coding agent works under the hood, from its security architecture to its unreleased feature roadmap. Two major data exposures in one week also raise serious questions about Anthropic's release hygiene at a critical moment in its growth trajectory.
Sources:
🔗 https://kuber.studio/blog/AI/Claude-Code's-Entire-Source-Code-Got-Leaked-via-a-Sourcemap-in-npm,-Let's-Talk-About-it
🔗 https://x.com/kimmonismus/status/2038976848168735181?s=20
🔗 https://www.theregister.com/2026/03/31/anthropic_claude_code_source_code/
Get what you want from TV advertising
What you want from TV advertising: Full-screen, non-skippable ads on premium platforms.
What you get: "Your ad is on TV. Trust us."
Modern, performance-driven CTV gets your TV ads where you want with transparent placement, precision audience targeting, and measurable performance just like other digital channels.
TV doesn't have to be a black box anymore.
Daily traffic to Claude since the beginning of the year. Still increasing by a lot.
OpenAI's AI Solves (another) Erdős Problems. But this time a single model alone.
AI models have been cracking Erdős problems for months now, GPT-5.2 solved Problem #728 autonomously back in January, and over 100 problems have moved to "solved" since late 2025. But OpenAI just raised the bar again. A new paper on arXiv presents three elegant proofs to open Erdős problems in combinatorics and number theory, and the authors state that every proof was produced entirely by an unreleased internal model. The humans? They just cleaned up the writing.
What's new here isn't that AI solved Erdős problems, that train left the station months ago. It's the scope and the claim. Previous milestones involved human-AI collaboration, iterative prompting, or sophisticated literature search. This time, a single internal model delivered three finished, publishable proofs across different mathematical domains without human mathematical guidance. Mehtaab Sawhney, one of the paper's authors who left Columbia for OpenAI, has predicted that AI will fundamentally change how mathematics is done. OpenAI also tested GPT 5.4 Pro on the same problems, and it independently solved two of the three in fewer than ten attempts.
Terence Tao has noted that AI systems are well-suited for the "long tail" of obscure problems where standard techniques apply but no human had systematically attempted them. The question now: how far up from that long tail can these models climb?
The best HR advice comes from those in the trenches. That’s what this is: real-world HR insights delivered in a newsletter from Hebba Youssef, a Chief People Officer who’s been there. Practical, real strategies with a dash of humor. Because HR shouldn’t be thankless—and you shouldn’t be alone in it.