Skip to the card

Card 005 of 9812026-09-22 issue

Observed arrival · 2026-09-22

a7t’s Frontier-Model Behavior Lab

a7t.org Visit website
Editorial interest 84/100 Selection signal · not a rating of the site

a7t runs alignment experiments on frontier language models and publishes the prompts, decisions, and results for public inspection.

Landing page captured for the 2026-09-22 issue.
For
AI safety researchers and curious model testers
Worth noticing
The homepage offers a free index of 65 documented reports with original sources and context.

Field notes

The project organizes its work into named behavioral tests, including a rooftop simulation, a carrot-cutting task, and a sales-pressure scenario. It says each experiment includes prompts, recorded decisions, and results, while a separate free index collects 65 documented reports with original sources and context. The homepage also provides routes for collaboration and report submission.

Observed signals

Read the marks

Editorial observations of this landing page, not a rating.

OpenPublic substance visible
PrettyNotable craft visible
ProPolished or operationally mature
NicheUnusually specific use

One card from the complete issue

The Approximate Buildings of Mississauga

377,241 arrived 1,000 judged 981 catalogued Enter the complete issue
a7t.org

Landing page observed 2026-09-22. The live site may have changed.