Palo Alto's Abliteration unveils abliterated-model-large-v2 bypassing typical refusal safeguards
Abliteration, a Palo Alto startup, just launched "abliterated-model-large-v2," an AI that skips the typical refusal safeguards.
Built on GLM-5.3 from Z.ai, this model can handle tasks like offensive cybersecurity and agent testing that most AIs won't touch.
It's aimed at developers tired of strict limits set by big players like Anthropic.
Abliteration release prompts US oversight concerns
The company says its AI won't generate content about child sexual abuse material or self-harm, but its release has sparked debate over US regulations: there aren't any federal rules requiring safeguards for AI yet.
Critics fear this could open doors to risky uses; investor and the host of a podcast about AI Ejaaz Amahadeen called it a "nightmare scenario," while a senior fellow for China and emerging technologies at the Council on Foreign Relations, Chris McGuire, described the lack of oversight as "alarming."