Extreme scale AI models are expensive to train and only a few tech companies can afford development, concentrating power while making it impossible for broader research community to inspect, dissect, and audit these systems for safety, creating a governance and epistemic crisis