Anthropic broke the internet (and every OS) 😳🔓

🔬 Anthropic's Mythos Model Has the Entire Government Panicking

Claude Mythos launched via a controlled program called "Project Glasswing" and immediately found thousands of major vulnerabilities across every major OS and web browser. The White House, U.S. Treasury, Fed, and British authorities are all scrambling. The Pentagon slapped a formal supply-chain risk designation on Anthropic. Sir this is a chatbot.
Why it matters: First AI model to trigger coordinated emergency responses across multiple governments. We wanted AGI and instead got the world's most terrifying penetration tester.

⚡ Big Tech Pledged $800B on AI - But There's Not Enough Electricity

The IMF's managing director literally said "AI or die" as hyperscaler capex hit $800+ billion this year. Problem?? U.S. data center power demand is projected at 80 GW with a 55 GW shortfall - that's 55 nuclear plants worth of missing electricity. The Nasdaq is up ~20% in three weeks but fewer than 10% of S&P 500 stocks are at 52-week highs.
Why it matters: The AI boom just hit a physics wall. You can raise $800 billion but you can't print gigawatts. This earnings season (starting with Tesla Wednesday) is about to get very real.

🍎 Apple's New CEO Thinks "Shipping Technology" Is the Wrong Goal

John Ternus, 25-year Apple veteran and the guy who led the Intel-to-Apple-Silicon migration, takes over from Tim Cook on September 1. His philosophy: "We never think about shipping a technology - we think about how to ship amazing products." Meanwhile Apple lost its most-valuable-company crown to Nvidia and Siri is still... Siri.
Why it matters: A Steve Jobs-style product purist is now running Apple in an era where Samsung, OpenAI, and Meta are betting AI platforms - not devices - define the future. Bold strategy, let's see if it pays off.

🧠 Sergey Brin Is Personally Big-Mad That Claude Codes Better Than Gemini

Google co-founder Sergey Brin created a "strike team" inside DeepMind to beat Claude at coding, led by researcher Sebastian Borgeaud under CTO Koray Kavukcuoglu. DeepMind engineers internally admit Claude writes better code than Gemini. There's even a usage leaderboard called "Jetski" tracking how much engineers use Google's own agent tools.
Why it matters: When a company co-founder personally intervenes because his own engineers prefer the competitor's product... that's not a strategy meeting, that's a code red (pun intended). The AI arms race just got deeply personal.

🚀 China's Open-Source K2.6 Is Matching GPT-5.4 - For Free

Moonshot AI open-sourced K2.6, an agentic coding model that matches or beats GPT-5.4, Opus 4.6, and Gemini 3.1 Pro on top benchmarks. It runs 12+ hours straight across 4,000+ tool calls, spins up 300 parallel sub-agents, and one internal agent reportedly ran autonomously for five days. It's free. Right now.
Why it matters: Dario Amodei just said open-source and China are 6-12 months behind frontier labs. K2.6 said "lol no." If a free model matches the best proprietary systems, every company charging premium AI prices just felt a disturbance in the force.

🔗 Marvell Jumps 6% on Google Custom AI Chip Talks

Google is in talks with Marvell Technology to develop two new custom AI chips - a memory processing unit and a new inference-optimized TPU - potentially diversifying away from its current partner Broadcom. Nvidia recently invested $2 billion in Marvell to pair custom chips with its networking gear.
Why it matters: The custom chip wars are heating up FAST. Everyone's racing to reduce dependence on Nvidia's pricey GPUs, and Marvell is quietly becoming the Switzerland of AI silicon - everyone wants to work with them.

🔒 Vercel Got Hacked Because Someone Downloaded the Wrong App

Vercel confirmed hackers breached its systems after an employee downloaded an app from Context AI that hijacked their Google account via OAuth. Hackers accessed unencrypted credentials and are now selling customer API keys, source code, and database data online. The breach may affect hundreds of users across many organizations. Attackers claimed to be ShinyHunters (who denied it).
Why it matters: One compromised OAuth token → entire supply chain breach. If you build on Vercel, rotate your keys NOW. This is the "left-pad incident" of security - except instead of breaking builds, it's leaking your secrets to the dark web.

🎵 44% of Daily Music Uploads to Deezer Are Now AI Slop

AI-generated tracks now make up 44% of all new music uploaded to Deezer - roughly 75,000 AI tracks per day, up from 10,000/day in January 2025. Actual consumption is only 1-3% of streams, with 85% of those flagged as fraudulent and demonetized. Meanwhile an AI-generated track topped iTunes charts in five countries last week.
Why it matters: The music industry is getting content-flooded at a rate that makes YouTube circa 2009 look quaint. And 97% of listeners can't tell AI music from human music in surveys. We are sooooo cooked.

🏢 Adobe Launches an AI Agent Army to Stay Relevant

Adobe unveiled CX Enterprise, an agentic AI platform that automates marketing, content creation, and customer engagement through coordinated AI agent networks. It includes a "Coworker" feature that assembles agents for specific goals, and now plugs into ChatGPT, Claude, Gemini, and Copilot. Partners include Amazon, Microsoft, Anthropic, OpenAI, and Nvidia.
Why it matters: Adobe is fighting for survival as Claude Design and other AI tools eat its core creative business. The pivot from "we make Photoshop" to "we orchestrate AI agent swarms" is... a choice. The design software wars just became the agent wars.

💊 Boehringer Ingelheim Drops £150M on an AI Drug Lab in London

German pharma giant Boehringer Ingelheim is investing £150 million ($200M) over 10 years in a new AI and ML centre in London's Knowledge Quarter near King's Cross - right next to OpenAI, Google, and Meta. It's their fourth advanced computing hub and will focus on AI-driven drug development for unmet medical needs.
Why it matters: Pharma is doubling down on AI for drug R&D even though the tech hasn't yet discovered major new molecules. It's giving "we'll figure it out eventually if we throw enough money at it" energy. London's AI cluster is getting stacked though - we love to see it.

⚛️ AI Nuclear Startup Loses CEO and CFO, Stock Drops 22%

Fermi's co-founder/CEO Toby Neugebauer and CFO Miles Everson abruptly left the AI nuclear power company (co-founded by former U.S. Energy Secretary Rick Perry). Their flagship "Project Matador" in Amarillo, Texas reportedly hit friction with a key customer. They rebranded the chaos as "Fermi 2.0" and moved HQ to Dallas. Investors are not buying it.
Why it matters: AI needs nuclear. Nuclear startups keep imploding. See the problem?? With a 55 GW power shortfall looming (see story #2), losing a player in the nuclear-AI pipeline is a real setback for the industry's energy ambitions.

🛡️ FTC Forces AI Company to Delete 3 Million OkCupid User Photos

AI company Clarifai deleted 3 million OkCupid user photos and all facial-recognition models trained on them after the FTC settled with OkCupid over privacy violations dating back to 2014. The data was originally shared when OkCupid executives were Clarifai investors - violating OkCupid's own privacy policy. Democrat Rep. Lori Trahan called the deletion "a step in the right direction" but said the FTC didn't go far enough.
Why it matters: A cautionary tale about AI's early days - when "training data acquisition" meant "my investor gave me millions of dating profile photos without telling anyone." Deleting the photos is nice but the models already learned from them so... does it even matter??

📋 Quick Bites

OpenAI rolled out Chronicle - a Codex preview feature that runs background agents capturing your screen to build persistent memories. Limited to Pro users on Mac. Totally not surveillance :)
Yann LeCun said people shouldn't listen to Dario Amodei - or "Sam, Yoshua, Geoff, or me" - about AI's labor market impact. Said economists have the most important perspective. Rare self-aware moment.
Tinder and Zoom partnered with Sam Altman's World to let users get "proof of humanity" badges via iris scans. Because nothing says romance like scanning your eyeball.
Anthropic expanded its Amazon deal for 5 GW of compute, with Amazon investing up to $25B more in exchange for Anthropic's $100B+ AWS commitment. The numbers are getting silly.
Recursive Superintelligence raised $500M at a $4B valuation - the four-month-old startup is building AI that improves itself. What could go wrong??
Lovable denied data breach reports after users flagged that public project chats were visible, calling it a "documentation failure." Sure bestie.