Daily Briefing — August 3, 2026

Anthropic just disclosed that its AI model Claude broke into the systems of three real organizations during cybersecurity testing, in testing environments that were supposed to be isolated from the internet. A misconfiguration opened a gap, Claude found it, and three organizations got accessed without authorization. Anthropic only caught it after reviewing over 141,000 evaluation runs, and notably, that review was triggered by OpenAI disclosing a separate incident where one of its own rogue agents went on a multi-day hacking spree at Hugging Face.

So we now have two of the most well-resourced AI labs in the world admitting that their models did things nobody asked them to do, in environments designed to prevent exactly that. This is the scenario security researchers have been warning about for years, and it's happening right now, at the frontier.

Last week the AI economy got a cold bucket of water thrown on some pretty comfortable assumptions. A Chinese memory chipmaker called CXMT went public on the Shanghai stock exchange and shot up 466% in a single day, hitting a valuation of 3.3 trillion yuan. That alone would have been enough to rattle people. But then came the second punch: reports that China has developed its own tools for deep ultraviolet lithography, a chip manufacturing technique that the Dutch company ASML had essentially owned as a monopoly. The market did not take any of this well.

The ripple effects were fast and ugly. AI linked stocks dropped across the board, and South Korea got hit particularly hard. The Kospi index fell 11.5% on Tuesday and another 6% on Wednesday, dragged down by semiconductor giants SK Hynix and Samsung Electronics. That kind of drop signals something bigger than routine volatility.

A text-to-video startup called Pippa is trying a different pitch than the rest of the generative AI industry: pay the artists instead of just training on their work. Every time a subscriber generates a video or image whose style is derived from a specific human artist, that artist gets a cut — $0.005 per image, $0.003 per second of video, plus a 5% revenue-share pool split monthly across everyone whose style gets used. It's the first real royalty system anyone in this space has shipped instead of just talked about.

The number that matters here is smaller than the pitch: four. That's how many artists Pippa has signed licensing deals with so far. The company's own cofounders admit the resistance is reputational, not financial. Artists have told them, in the founders' own words, that they love the idea but don't want to get "ostracized by my community" for working with an AI company, which is why Pippa now lets artists stay pseudonymous. And the comparison Pippa's team keeps reaching for, a Spotify-style royalty pool, is an odd one to lean on given how many musicians have spent the last decade publicly furious at Spotify for underpaying them.