Abliterated models terrify me, but so do safety filters: paraschopra on AI's fairness asymmetry
paraschopra · x · 2026-10-28
paraschopra lays out a security dilemma: abliterated models with refusals deleted can enable cyberattacks or bioweapon design, yet mainstream safety filters are so aggressive that Claude degrades you to an older model for any bio- or security-adjacent question.
His key points:
- If models keep refusing uncharted research directions, how can ordinary scientists do frontier work? Anthropic/OpenAI internally use versions that refuse less — an unfair asymmetry.
- Attackers will use abliterated models regardless, so how do defenders prepare?
His conclusion: either ban abliterated models for everyone or give everyone equal access — the current asymmetry is untenable.
More from Models
- DeepSeek vs Haiku pricing debunk: accounting for peak rates and token use, 1017 is 2-2.5x cheaper — teortaxesTex · 2026-10-28
- ChatGPT and Gemini refuse to draft a complaint against a pedophile, user reports — chinchinlover-419 · 2026-10-28
- GPT-6.1 Terra priced at $1/$5 per M tokens, but the '9x cheaper per task' math doesn't hold — 2C_ornot2C · 2026-10-28
- FrontierCode 2 leaderboard sparks drama over missing Chinese models — teortaxesTex · 2026-10-28
- Evil 27B Too Risky to Run Autonomous? User Fears Wiped Drive Over a Game — teortaxesTex · 2026-10-28
- DeepSeek V5 rumor debunked: code strings were just September assert tests — teortaxesTex · 2026-10-28