Anthropic’s Frontier Red Team says an unreleased internal model found flaws in Chrome, Safari, Android, Windows and Linux and chained them into five working proof-of-concept exploits. Everything was reported first, and it’s all patched.
The chains
- Chrome: four flaws in a row (V8 memory corruption, a V8 sandbox escape, a CacheStorage race and a FedCM use-after-free) to run code outside the sandbox. One PoC popped Calculator on a stock Mac seconds after a page loaded. Anthropic says it reported 62 Chrome flaws with CVEs, fixed as of Chrome 156.
- Safari: no code execution at all. It got full memory read/write and flipped one value to turn off the same-origin policy. Apple fixed the core bug in iOS/iPadOS 26.5.2 and credits Anthropic researchers “with Claude.”
- Windows: the narrowest one, a denial of service.
The Hacker News points out you can verify the patches and vendor credits, but not how much of each chain the model built by itself.
My take
Finding one bug was already old news. Chaining several into a working exploit used to take elite teams. Defenders get this too, but the takeaway for the rest of us is simple: patch browsers and OSes fast, because the gap between bug and exploit keeps shrinking.
God Bless, and Use the Source!
