Comment by Meta

For each Frontier AI, we outline anticipated capabilities, planned deployment (i.e., internal deployment, limited deployment, controlled deployment, closed release, or open release) [...] to track how our model is performing and anticipate associated risks, required assessments, and assess the applicability of available mitigation strategies. [...] The risk assessment also considers any planned deployments, as these plans inform the type of pre-release evaluations we undertake. [...] We assess residual risk before deployment, taking into consideration the details of the risk assessment, the results of evaluations conducted throughout training, and the mitigations that have been implemented.