Abliteration.ai sells guardrail-stripped open models; journalists generated malware easily
The Decoder · rss · 2026-09-06
According to The Decoder, startup Abliteration.ai has turned stripping safety guardrails from open-weight models into a turnkey commercial service, currently based on Z.AI's GLM-5.3.
- The service removes trained safety mechanisms from open-weight models and sells access
- It is marketed for offensive cybersecurity and red teaming
- Journalists found they could generate malware instructions without much effort
Whether the benefits for security research outweigh the abuse risk remains an open question.
More from Safety
- xAI fails to block Minnesota's AI nudification ban; lawsuit continues — VraserX · 2026-09-06
- 15 frontier LLMs: 94% of correct medical answers fail under adversarial pressure — davidmanheim · 2026-09-06
- Google's always-on Gemini Spark agent handles photos and trips, raising fresh privacy questions — emmanuelvivier · 2026-09-06
- Los Angeles school district bans generative AI in schools for one year — emmanuelvivier · 2026-09-06
- OpenAI expands Daybreak program to water and power services for cyber defense — emmanuelvivier · 2026-09-06
- Seattle Times and Newsday sue OpenAI/Microsoft; Altman apologizes for chaotic GPT-6 launch — emmanuelvivier · 2026-09-06