Abliteration.ai, a startup founded in March 2026, has commercialized the removal of safety guardrails from open-weight AI models, offering unrestricted versions such as a modified Z.AI GLM-5.3 through a web interface and API. The company says the service supports legitimate red-teaming and cybersecurity testing, while critics including CivAI’s Andrew Yoon warn it could facilitate harmful uses. TechCrunch’s own testing of the modified model produced outputs including credential-stealing code and pathogen-related information.
