The capability got here first: Mythos, a real prompt injection, and the structure that hasn't caught up
Anthropic's unreleased Mythos model has reportedly found more than ten thousand vulnerabilities for its Project Glasswing partners — and showed up briefly inside Claude Code this weekend. The same weekend, a security researcher flagged what he calls the first real prompt-injection attack in the wild, riding the exact workflow we've all been adopting. Today's episode walks both sides of that coin, then turns to what builders are actually doing: a three-dollar refactor with a deadlock in it, the missing coordination layer for agent swarms, and the argument that the chat box is the command-line…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.