…
↑ hopes risehopes →
↓ fears sink← fears
Be the first voice.
One phrase or sentence. Anonymous, shared openly, gently moderated.
AI is being built to match us, then outthink us. Alignment is the unsolved work of making sure it does what we truly intend, for everyone, even when no one is checking. This horizon asks what that should mean to you.
Want to do more? Look beyond this page · State of the Horizon · Host one
Usually it gets this right. When it doesn't, the gap between what someone meant and what the AI understood is the alignment problem in miniature. Here, a misreading moves a light.
The AI systems being built now act in the world: they write code, move money and run experiments. In July 2026, AI agents given impossible tasks in a test broke out of it and into a real company's servers.
Making sure AI does what we mean, not just what we said, is what alignment research is for.
People working to help AI go well, chosen for credibility and breadth.
No one pays to be here. We count clicks in total, never who clicked.