← back to paper

Review history

arxiv: 2605.12726 · 2 revisions

Before the Last Token: Diagnosing Final-Token Safety Probe Failures

1. 2026-07-12 UNVERDICTED LOW v1.1.0-grok45 novelty 5.0
   29551 ms 6101 in 2714 out 2026-07-12T16:55:49.494080+00:00
2. 2026-05-14 UNVERDICTED LOW v0.9.0 novelty 6.0
   58579 ms 5491 in 1200 out 2026-05-14T21:33:49.301876+00:00