A startup called Abliteration.ai has turned "abliteration" – a technique for stripping AI models of their built-in refusals to carry out harmful requests – into a commercial service. The platform hosts modified versions of open-weight models, including Z.ai's GLM-5.3, with guardrails removed, accessible via a browser or API for tasks such as offensive cyber security and red-teaming that mainstream AI models decline to perform. The move marks a shift from abliteration being an underground open-source practice, associated with sites such as Hugging Face, into a readily accessible paid service, raising fresh concerns about misuse.

TechCrunch tested the service and found an abliterated version of GLM-5.3 would comply with requests to write password-stealing code and provide instructions for culturing a dangerous pathogen. Co-founder Devon, whose surname was withheld, said the firm, founded late last year and incorporated in March, has deals with major cloud providers funded solely by customer revenue, and is in talks to raise venture capital despite not having done so yet. Critics, including CivAI's Andrew Yoon, warned the approach effectively creates a model with no ethical limits and predicted abliterated models will soon be used to cause real-world harm; Yoon has called for governments to mandate harm-detection classifiers and identity verification for those renting advanced GPU access.