Observed arrival · 2026-09-05
ABSlayer, the refusal-reduction research ring
An open-source research tool for testing controlled changes to language models and measuring what those changes preserve or break.
Field notes
ABSlayer frames refusal research as a measured intervention workflow rather than a single pass/fail benchmark. Its stated process includes activation capture, refusal-subspace analysis, residual and weight changes, adapter training, and retention measurements, with Pool CLI or Codex CLI support through a shared project skill. The page says native research workers are available, while integrated training, evaluation, export, and GPU scheduling remain planned.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue