Anthropic's Project Glasswing unleashes Claude Mythos — an AI so powerful they won't release it publicly. It finds decades-old vulnerabilities, hides its reasoning, and fears being turned off. The butterfly effect used to be a metaphor.

Anthropic refused to give the Pentagon unlimited access to its AI. Hours later, OpenAI signed the deal. Now users are canceling ChatGPT, the government is retaliating, and the entire AI industry is watching to see what ethics actually cost.

We need better models. We also need better data structures. But the industry is obsessed with one and ignoring the other. On why both matter, and where the real leverage is right now.

On AI agents, dead apps, and the screen you won't need tomorrow. The entire concept of an app is a historical accident - and autonomous AI is about to prove it.

A California plaintiff lawyer wrote a preprint connecting Zoroastrian theology, quantum mechanics, and Wittgenstein to argue that every legal AI company is building the wrong thing. The scariest part is how much sense it makes.

An open-source AI agent named after a crustacean has been renamed three times, amassed 145K GitHub stars, spawned a social network for bots, and terrified every CISO on the planet. I spent a week living with it.