Startup offers unguarded AI models as a commercial service for red-team testing

Abliteration.ai, founded late last year and incorporated in March, commercializes the practice of removing safety guardrails from open-weight AI models, offering versions like Z.ai’s GLM-5.3 through a web interface and API. The startup argues that unrestricted models let security teams reproduce malicious tactics for better defense, and it already serves early-stage red-team firms in the UK and Europe. Critics, including Andrew Yoon of CivAI, say the service could turn models into “sociopaths” and facilitate cyber or bioweapon misuse.

While the platform imposes a few minor limits—such as refusing suicide instructions—it lacks comprehensive KYC and relies only on credit-card logging. Abliteration.ai finances its operations through customer fees, has partnerships with major cloud providers, and is in early talks for venture funding.