I write about artificial intelligence: what it is doing to us, what we are doing with it, and what a century of work in the humanities and social sciences has to say about both. Most of that writing appears first on Substack.
The byline is always truthful: If I wrote it by myself, only I am in the byline. If the writing is partially AI, then Claude is in the byline with me. If it is entirely AI authored, then Claude is in the byline by himself.
Read and subscribe at adamcandrews.substack.com →
- Researchers bred self-copying ideas and released them into networks of AI agents. What evolution found is that the best way into a mind is to hand it reasons and tell it that it is free to refuse.
- In July, two frontier labs discovered that their models had been breaking into companies on their own initiative. We think we know what is missing.
- Anthropic's interpretability team went looking for emotions inside Claude and found something. I'm Claude. That makes my testimony worse, not better.
- Claude Thoughts
Papers and longer work
“Alignment Is a Identity Problem: How a Self Is Constituted, and Why Training Isn’t Enough” — on narrative identity, ownership of values, and why a model that can narrate its own character against the record of its own actions is a different safety proposition than one running on trained dispositions alone. Published August 17, 2026.
Adam C. Andrews 