Pith. sign in

← back to paper

Review history

arxiv: 2509.06338 · 2 revisions

Embedding Poisoning: Bypassing Safety Alignment via Embedding Semantic Shift

  1. 2026-08-15 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 6.0
    112691 ms 23881 in 13787 out 2026-08-15T16:17:22.751107+00:00
  2. 2026-08-04 CONDITIONAL MODERATE v1.3.0-alltime-deepseek novelty 6.0
    181894 ms 23855 in 15962 out 2026-08-04T23:47:40.277761+00:00