Abliteration CEO Devon Thomas argues that reduced-refusal models can help security teams test their own systems and AI agentsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. The interviewer challenges him about potential misuse and the limits of safeguardsAI safety is the field and practice of reducing harmful failures, misuse, loss of control, and unintended consequences from AI systems..
Thomas discusses model quality trade-offs, access controls and customer-defined guardrailsA guardrail is a technical or procedural control that limits unsafe, disallowed, deceptive, or otherwise unwanted AI behavior.. His claims about security and the effect of refusals are presented as his views, rather than independent findings.
Watch on YouTube
