Cryptographer Matthew Green: can abliterated GLM write 0days? Just test it
matthew_d_green · x · 2026-11-06
Cryptographer Matthew Green raises an open question: can an "abliterated" GLM (with safety refusals removed) actually produce 0day exploits? He notes the question seems answerable by simply spinning up a cloud instance and trying it — suggesting empirical testing over speculation. The post touches on the real offensive capability of uncensored open-weight models.
More from Models
- Menlo report: Enterprise AI spend soars to $177B, up 5.4x in a year as agents reshape buying — mmurph · 2026-11-06
- vLLM releases Kimi K3.1 Pro pruned 75%: 896→224 experts per MoE layer, ~1.5TB down to ~470GB — victormustar · 2026-11-06
- Tom Goldstein teases more sphere models coming very soon, with Meta AI collaborators — tomgoldsteincs · 2026-11-06
- GPT-6-Sol's unscored RCE chain: fixing a scorer bug ties it with GLM-5.3-Flash — teortaxesTex · 2026-11-06
- Same 10 prompts, one year apart: GPT-6 output dwarfs the GPT-5 3D editor demo — Just_Lingonberry_352 · 2026-11-06
- First recorded case: lone operator runs 12 AI agents to breach orgs in 11 countries — Jsevillamol · 2026-11-06