Experts Urge Periodic Internal AI Model Evaluations for Loss-of-Control Risks

At least for loss-of-control risk, the right timing is periodic evals of the most capable internal models, rather than evaluating a model right before public release