The document, which Meta is calling its Frontier AI Framework ... security protections to prevent the system from being exfiltrated and stop development until the system can be made less dangerous.