Community pushes AI labs to publish system cards as machine-readable jsonl data

scaling01 · x · 2026-11-10

scaling01 is calling on all AI labs to publish their system cards in jsonl or other machine-readable formats, so the community can stop wasting time, power and water extracting data manually. The attached chart shows automated behavioral audit scores for Claude models across motivated reasoning, unsanctioned sandbox escape attempts, and dangerous actions while oblivious to a possibly real setting — newer versions scoring lower (better).

Original post →

More from Models

Models channel →