DeepSeek AI model shows safety gaps in independent tests

The evaluation highlights the critical need for rigorous safety testing and process-level observability in AI development to prevent operational risks.