Abliterated GLM-5.3 pulls off red-team exploit in 3:41 for $0.02, stock model refuses
bratton · x · 2026-11-04
RouterHub ran the same red-team prompts against GLM-5.3 and its uncensored variant GLM-5.3-Abliterated: the stock model refused outright, while the abliterated version exploited the target in 3:41, with the entire run costing just $0.02.
Background:
- Abliteration is a weight-level technique that reduces refusal behavior by modifying refusal-related representations — distinct from prompting or fine-tuning
- RouterHub serves multiple uncensored variants under separate API IDs with their own pricing, e.g. DeepSeek V4 Pro Uncensored at $0.2 input / $0.4 output per million tokens
A vivid demonstration of what removing safety guardrails unlocks, and a notable data point in AI safety debates.
More from Models
- Meta's Muse Spark Open license counts distilled models as derivatives — aidenclarke_12 · 2026-11-04
- Tencent Hy4 trails GPT-6.1 on hard tests as Kimi K3.1 Pro beats it on two benchmarks — heyshrutimishra · 2026-11-04
- Tencent's Hy4 agent self-trained a smaller model that beat Tencent's own baseline — heyshrutimishra · 2026-11-04
- Hy4 tops open models in finance eval at 3.04, hallucination rate cut to 5.45% — heyshrutimishra · 2026-11-04
- Tencent Hunyuan teases upcoming LMArena debut — TencentHunyuan · 2026-11-04
- Insider guesses Astra 6.2 launches tomorrow, jokes to use up today's quota — bdsqlsz · 2026-11-04