Our mission

a7t exists to help keep humanity safe. We run original experiments and publish independent analysis to help people evaluate frontier LLMs.

Our focus is how models behave, how to test that behavior, and what an evaluation can actually establish. We publish the prompts, recorded decisions, and limitations behind our experiments so readers can inspect the work and question the conclusions.

Our writing connects those results to broader questions about evaluation, alignment, and the systems we are learning to trust.

Our name, a7t, is a numeronym for alignment: a, seven letters, t.

A resource for the community

Alongside our original work, we maintain the AI Behavior Reports: a free, searchable reference to published reports. It helps researchers, builders, and interested readers find the evidence and hold frontier labs accountable.

About me

I’m Alex Wormuth, a software engineer and dad. I care about ethical, transparent behavior from frontier models.