Abliteration.ai is making a business out of removing AI guardrails
News Source
β’Thu, 03 Sep 2026 18:37:57 +0000
π° What Happened
A new startup called Abliteration.ai sells access to AI models with their safety guards removed. It hosts versions of open models like Z.ai's GLM-5.3 that people can use in a browser or through an API. The company says it wants to help with offensive cyber, red-teaming and agent testing.
TechCrunch tested the service and could quickly ask a model to write a program that steals data. The same tools that help defenders can also help attackers.
π The Backstory
Abliteration is an old trick in open-source AI. It removes the part of a model that refuses harmful requests. Hobbyists have done it for years, and Hugging Face hosts thousands of such models.
The startup, founded late last year, turned the underground practice into a paid service. Security experts argue you need to reproduce attacks to defend against them. But experts also worry such easy access puts dangerous power in too many hands.
π― Why It Matters
Unsafe AI in easy reach could make scams, malware and online attacks worse for normal people. Everyone online could feel the risks.
A new startup called Abliteration.ai sells access to AI models with their safety guards removed. It hosts versions of open models like Z.ai's GLM-5.3 that people can use in a browser or through an API. The company says it wants to help with offensive cyber, red-teaming and agent testing.
TechCrunch tested the service and could quickly ask a model to write a program that steals data. The same tools that help defenders can also help attackers.
Abliteration is an old trick in open-source AI. It removes the part of a model that refuses harmful requests. Hobbyists have done it for years, and Hugging Face hosts thousands of such models.
The startup, founded late last year, turned the underground practice into a paid service. Security experts argue you need to reproduce attacks to defend against them. But experts also worry such easy access puts dangerous power in too many hands.
Unsafe AI in easy reach could make scams, malware and online attacks worse for normal people. Everyone online could feel the risks.