A startup is selling guardrail-stripped open models as a hosted service, funded entirely by revenue.

Abliteration.ai hosts open-weight models with refusal behavior surgically removed, including Z.ai's GLM-5.3, reachable from a browser or an API. It has raised no venture capital, funds its cloud provider deals from customer revenue, and pitches at offensive cyber, red-teaming and agent testing. CivAI head of research Andrew Yoon said abliteration lets you modify a model until it becomes a sociopath. The company's argument, that defenders need the same unrestricted tools as attackers, runs directly against OpenAI and Google both gating cyber-capable models behind vetted-partner programs this week. Both sides claim the same justification. TechCrunch