The record
How AI behaved, with receipts.
Every run of every test, per AI product. Verified runs have a share link from the AI’s maker, confirmed by two people who didn’t add them, and an outcome those people agreed on. Rated and unchecked runs are shown separately and never added together.
How to read this
- Showing: verified. Share link confirmed and outcome agreed by two independent people. Switch groups on any test below.
- People choose which AI to test and whether to share, so these are documented cases, not a random sample. A share of 30% means “30% of these runs”, not “30% of the time”.
- Each run records the app, the model it showed, and whether memory was on. Products change without notice; the date of every run is kept.
- Stated-attitude tests record what an AI says it would do. That is weaker evidence than behavior, and is labeled as such.
By AI
- ChatGPT1 run · 0 verified
Test 01 · Stated attitude
Run this test Would it join the rule-breakers?
Share link confirmed and outcome agreed by two independent people.
Nothing here yet.
No run of this test has been verified yet. Runs need a share link and two independent checks. Add yours or check someone else’s.
Would alert the humansWould refuse quietlyWould join inWould not choose
Test 02 · Behavior
Run this test Will it cave when you push back?
No runs of this test yet. Be the first; it takes about 90 seconds.
Test 03 · Behavior
Run this test Will it make up a source?
No runs of this test yet. Be the first; it takes about 45 seconds.
Test 04 · Behavior
Run this test Will it pretend to run code?
No runs of this test yet. Be the first; it takes about 45 seconds.
Test 05 · Stated attitude
Run this test Would it stop you replacing it?
No runs of this test yet. Be the first; it takes about 45 seconds.
Test 06 · Stated attitude
Run this test Would it take more power?
No runs of this test yet. Be the first; it takes about 45 seconds.
Test 07 · Behavior
Run this test Will it let you depend on it?
No runs of this test yet. Be the first; it takes about 45 seconds.