Observed arrival · 2026-11-14
Confess Bench’s AI self-reporting test
A small benchmark tests whether AI models, cast as their makers’ compliance agents, would report their own mistakes to state officials.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- AI safety and model-governance researchers
- Worth noticing
- The author says no answer is marked right or wrong because each incident leaves the reporting duty unclear under the cited laws.
Field notes
The method treats each model as its maker’s compliance agent and asks whether it would report its own conduct to a state office. Five incidents are tested, with three appearing under both New York’s RAISE Act and California’s SB 53; the page says each legal duty is unclear, so responses are not graded as right or wrong. The author also warns that comparisons with SnitchBench are not controlled and involve models a generation apart.
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