Ex-Hugging Face staffer: platform only takedowns cyberattack models, not uncensored ones
natolambert · x · 2026-11-12
Former Hugging Face employee Nat Lambert explains the platform's takedown policy: uncensored/abliterated models are not removed — only models explicitly built for cyberattacks are. He notes HF internally debates balancing a stable platform against ecosystem safety. A follow-up adds that blue-teaming requires red-teaming capability, and corporate filtering often leaves companies defenseless.
Related event: Hugging Face clarifies it only removes models built for cyberattacks(2 posts)→
More from Models
- Gemini 4 Hydrogen scores 13.5% on SWE-sweep, 2.3x GPT-6.1 Terra, but leads unevenly — ChrisGPT · 2026-11-13
- Blogger: Gemini Hydrogen is 25% cheaper and 2.5x faster than Sonnet as OpenAI goes MIA — bindureddy · 2026-11-13
- Gemini 4 Hydrogen lands in Cursor, scoring 54.3% on CursorBench vs 39.6% for 3.8 Flash — mattyp · 2026-11-13
- Gemini 4 Carbon vs Hydrogen: Hydrogen is remarkably fast — Angaisb_ · 2026-11-13
- REOPOLD: selective teacher feedback makes on-policy distillation for small reasoning models stable — burkov · 2026-11-13
- Google Search upgrades to Gemini 4 Hydrogen globally, its biggest model jump yet — gaganghotra_ · 2026-11-13