Anthropic red team's first-ever model "abliteration" cut GLM-5.3 refusals from 90% to 3% for ~$4,400
cephaloform · x · 2026-10-16
Anthropic's Frontier Red Team says it went two years without abliterating a single model — until GLM-5.3 two weeks ago. Per the report, the team produced an abliterated copy to study guardrail bypass, spending 2,200 GPU hours (600 for GLM-5.3-Flash) at a cost of roughly $4,400. The edit dropped the refusal rate from over 90% to about 3% and 2% on JailbreakBench and HarmBench respectively, illustrating how cheaply open-model safeguards can be stripped.
More from Models
- OpenAI ships GPT-Live-1.1: new voices in six languages, fewer confirmations — OpenAI · 2026-10-16
- Muse Spark model appears on LMArena, testable in Battle and Agent modes — arena · 2026-10-16
- Running a multi-model UI design arena: Paseo orchestrates four design drafts per model — vista8 · 2026-10-16
- Ilya Sutskever questions Meta license clause taxing MaaS providers above $20M revenue — ilyasu · 2026-10-16
- Muse-Spark open-weights drop: 816B-A42B MoE model gets 359GB MLX 3-bit build — ns123abc · 2026-10-16
- Midas flags inconsistent Claude limits: 33% weekly and 45% window usage despite $100 credits, yet out of API credits — TheMoonMidas · 2026-10-16