AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
OpenAI's $122B Record Round & Anthropic's Claude Code Leak
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's April 1st, 2026, and today we've got two massive stories that actually broke late yesterday OpenAI's record-shattering funding round and a detailed look at what Anthropic accidentally exposed in their clawed code source code leak. Let's start with the OpenAI news because the numbers here are genuinely staggering. The company just closed a$122 billion funding round at an$852 billion valuation. That's the largest funding round in Silicon Valley history by a lot. To put that in perspective, OpenAI is now worth more than most Fortune 500 companies, and it hasn't even gone public yet. The round was co-led by SoftBank and Andreessen Horowitz, with Amazon, Nvidia, Microsoft, and a bunch of other major investors participating. About$3 billion came from individual retail investors through bank channels, and the company will be included in ARC Invest ETFs, which is a pretty clear signal they're building toward an IPO. What's really interesting is what OpenAI disclosed in their announcement. They're now generating$2 billion in monthly revenue. They have 900 million weekly active users and 50 million paying subscribers. Business revenue now makes up 40% of their total, up from about 30% last year, and they say they're on track for parity with consumer revenue by the end of this year. They're explicitly calling themselves an AI super app, now making clear they want to own the primary interface for how people use AI. This whole announcement reads less like a typical funding press release and more like a draft S1 filing. Now, speaking of Anthropic, they've had a rough week. You might remember the Mythos model leak from a few days ago. Well, yesterday, someone discovered that Anthropic had accidentally shipped a source map file with their Claude Code NPM package, exposing the entire readable source code of the CLI tool. The package has since been pulled, but not before the code was widely mirrored and analyzed. The most significant discovery is something called Kairos, an unreleased autonomous agent mode that was hiding in the code. It includes a dream skill for what they call nightly memory distillation, GitHub webhook subscriptions, background demon workers, and cron scheduled refresh every five minutes. This is the scaffolding for an always-on background running agent, and it represents a major product roadmap reveal that competitors can now see. The leak also showed some fascinating technical measures. There's an anti-distillation system that injects fake tools into API requests to poison training data if someone's recording traffic to train a competing model. There's undercover mode that strips all traces of AI authorship from commits, meaning AI authored code in open source projects will have no indication an AI wrote it. There's frustration detection via a simple regex that catches swearing. And there's native client attestation implemented at the HTTP transport level, basically DRM for API calls, which is the technical enforcement behind their recent legal fight with open code. One more interesting tidbit a comment in the code revealed Anthropic was burning about 250,000 API calls per day on failed autocompact operations. They fixed it with three lines of code. That's the kind of inefficiency that becomes very visible when your source gets leaked. Alright, shifting to product news. Salesforce just announced a major AI overhaul for Slack with 30 new features. The big headline is an upgraded Slackbot with what they're calling reusable AI skills. You define a task once and can apply it in different contexts. So you could say, create a budget and Slackbot will pull relevant info from your Slack channels and connected apps, create a plan, and automatically schedule a meeting with the right people. It's now an MCP client, can transcribe and summarize meetings and can even monitor your desktop activities outside Slack to provide contextual suggestions. Salesforce is clearly trying to push Slack Beyond Messaging into being a core business process platform. On the model front, Cohere released Transcribe a new open source speech recognition model that's now number one on Hugging Face's OpenASR leaderboard with a 5.42% word error rate. It outperforms Whisperlarge V3, Eleven Labs Scribe, and other leading models. It's a 2 billion parameter conformer-based model that supports 14 languages and is available under an Apache 2 license. If you're building anything with speech to text, this is worth a look. Also worth noting, a company called Prism ML just launched what they're calling the first commercially viable 1-bit LLMs. Their 1-bit bonsai models have binary weights instead of the usual 1.6 bit or 4-bit precision. The 8 billion parameter model fits in just 1.15GB of memory. That's 14 times smaller than a standard 8B model while matching benchmark performance. They're claiming 8x faster inference and 5x better energy efficiency. This could be a big deal for running models on edge devices and smartphones. And finally, Yup AI, a startup that let users compare outputs from 800 different AI models, is shutting down after raising$33 million from a 16Z just last year. They had 1.3 million users but couldn't find product market fit. The founders said AI models improved so fast that consumer comparison became less relevant, and the industry is shifting toward hiring PhDs for reinforcement learning feedback rather than crowdsourced input. The future is not just models but agentic systems. Their CEO wrote That's it for today. See you tomorrow.